Michel is available for hire
Hire MichelMichel Meneses
Verified Expert in Engineering
Artificial Intelligence (AI) Developer
Campinas - State of São Paulo, Brazil
Toptal member since August 8, 2023
Bio
Michel is a senior AI/ML engineer with comprehensive experience leveraging the latest AI research into development across multiple domains. With a solid technical background in ML and software engineering, Michel has assisted enterprises in successfully prototyping, benchmarking, and developing cutting-edge AI systems used by millions worldwide while authoring innovative scientific papers and patents.
Portfolio
EdLight
A/B Testing, AI Research, Algorithms, AWS Lambda, Amazon SageMaker, Python 3...
Springer Nature
Artificial Intelligence (AI), Machine Learning, Deep Neural Networks (DNNs)...
Toradex
Python 3, Machine Learning, Artificial Intelligence (AI), Deep Learning...
Experience
- Artificial Intelligence (AI) - 8 years
- Machine Learning - 8 years
- Deep Learning - 6 years
- Python 3 - 6 years
- NumPy - 6 years
- TensorFlow - 5 years
- Computer Vision - 3 years
- Speech Recognition - 3 years
Preferred Environment
Linux, Python 3, TensorFlow, PyTorch, NumPy, Scikit-learn, Pandas, Amazon Web Services (AWS), Deep Learning
The most amazing...
...thing I’ve developed is the on-device wake-up word customization for the voice assistant of one of the world's leading smartphone enterprises, launched in 2023.
Work Experience
Head of Engineering
2024 - PRESENT
EdLight
- Spearheaded the development and national scaling of a proprietary AI-powered platform across 30+ U.S. school districts, delivering real-time pedagogical analysis of handwritten mathematics to over 15,000 K-12 students.
- Led the end-to-end AI project lifecycle, including strategic scoping, technical stack definition, and the product roadmap to scale mathematical diagnostics across the United States.
- Hired, managed, and mentored a team of machine learning engineers, establishing a high-performance engineering culture that accelerated the platform's development from prototype to production.
- Steered the team in the curation of large-scale proprietary datasets and the establishment of robust performance metrics, ensuring the AI models remained pedagogically accurate and technically superior.
- Implemented sophisticated, multi-step AI workflows based on cutting-edge Visual Language Models (VLMs), enabling the high-precision interpretation of non-standardized handwritten mathematical notation.
- Architected and implemented MLOps best practices to fully automate the AI workflow, streamlining the process from experimentation and testing to large-scale deployment and real-time production monitoring.
Technologies: A/B Testing, AI Research, Algorithms, AWS Lambda, Amazon SageMaker, Python 3, Git, GitHub, GitHub Actions, Notion, Deep Learning, PyTorch, Large Language Models (LLMs), Multimodal Models, Computer Vision, Clean Code, Clean Architecture, REST, Machine Learning Operations (MLOps), APIs, Unit Testing, Software Engineering, Kanban, Exploratory Data Analysis, Supervised Learning, Supervised Machine Learning, Hypothesis Testing, Team Management, R&D, FastAPI, Serverless, Serverless GPUs, AWS Serverless Application Model (SAM), SGLang, Multimodal GenAI, HubSpot API, Retrieval-augmented Generation (RAG), Fine-tuning, Datasets, Weights & Biases, Open-source LLMs, Data Labeling, Transformers, LangGraph, AI Agents, Agentic AI, Streamlit, LVM, Data Analysis, System Design, Website Data Scraping, AWS Command Line Interface (CLI), AWS Cloud Computing Services, Weave, Intercom, vLLM, Serverless Architecture
Volunteer Peer Reviewer
2022 - PRESENT
Springer Nature
- Served as a volunteer recurrent peer reviewer for the Complex & Intelligent Systems journal published by the Springer Nature Group.
- Reviewed and provided relevant insights on research papers related to the application of new machine learning algorithms across a wide range of complex applications, such as phishing website detection, pedestrian tracking, and contour detection.
- Helped industrial and academic researchers improve the robustness of their research methodologies and result presentations by providing constructive and highly detailed feedback on their submitted manuscripts.
- Reviewed the literature to get updated on the latest technical approaches often introduced by the manuscripts submitted to the journal, such as bio-inspired artificial networks, extreme learning machines, and tracking-by-detection frameworks.
Technologies: Artificial Intelligence (AI), Machine Learning, Deep Neural Networks (DNNs), Convolutional Neural Networks (CNNs), AI Research, Research, Literature Review, Technical Research, Computer Vision, Image Processing, Tracking, Object Tracking, Object Detection, Feature Analysis, Deep Learning, Metric Learning, AI Testing
Senior Machine Learning Consultant
2024 - 2025
Toradex
- Led the development of Toradex AI Chatbot, architecting a high-precision automation solution to resolve complex technical support queries for a global community of thousands of embedded systems engineers.
- Served as Strategic AI Advisor to the Global Marketing Officer, conducting a full technical audit of the AI roadmap and redefining the solution architecture to solve critical intent-matching gaps and ensure long-term scalability.
- Re-engineered the baseline system to increase the autonomous resolution rate from 5% to 10%, a 100% improvement in the bot’s ability to resolve intricate technical tickets in a single turn (zero-shot) without human follow-up.
- Architected and implemented high-performance RAG frameworks and MLOps pipelines, leveraging advanced vector databases to optimize the accuracy of LLM responses for complex hardware and software documentation.
- Directed engineering strategy and team mentorship, overseeing a distributed team of engineers while establishing automated CI/CD workflows that ensured 24/7 service availability and seamless integration with the Toradex community portal.
Technologies: Python 3, Machine Learning, Artificial Intelligence (AI), Deep Learning, Natural Language Processing (NLP), Chatbots, AI Chatbots, Large Language Models (LLMs), Software Development, Software Engineering, Clean Code, Data Cleaning, Object-oriented Programming (OOP), OpenAI, OpenAI GPT-4 API, Retrieval-augmented Generation (RAG), Web Scraping, Website Data Scraping, Cloud, Cloud Computing, AWS Command Line Interface (CLI), AWS Serverless Application Model (SAM), AWS Cloud Computing Services, AWS Lambda, Serverless, Serverless Framework, Serverless Architecture, CI/CD Pipelines, GitHub, GitHub Actions, Git, Debugging, Linux, Debian, Debian Linux, Pinecone, AI Website Builders, LVM, Data Analysis, System Design, GraphRAG, Vector Databases
Lead Machine Learning Consultant
2024 - 2024
EdLight
- Led the development of the AI system used by EdLight to automatically evaluate thousands of images of student work submitted weekly by several schools in the US.
- Designed and executed the AI technical roadmap while guiding the AI team in curating large-scale datasets, implementing state-of-the-art large visual models for few-shot learning multi-label classification, and deploying those as REST APIs.
- Adopted MLOps best practices to automate the entire AI development workflow, including experimentation, testing, deployment, and monitoring in production.
Technologies: Python, Machine Learning, Natural Language Processing (NLP), Computer Vision, Artificial Intelligence (AI), Team Leadership, Machine Learning Operations (MLOps), Google Cloud Platform (GCP), Azure, Amazon Web Services (AWS), Project Management, Large Language Models (LLMs), Deep Learning, LVM, Data Cleaning, Data Analysis, Notion, Kanban, Team Mentoring, Mentorship, Software Engineering, System Design, Clean Code, Clean Architecture, Data Preprocessing, OpenAI, OpenAI GPT-4 API, Amazon SageMaker, Multimodal Models, Deep Neural Networks (DNNs), A/B Testing, Hypothesis Testing, R, Pandas, Exploratory Data Analysis, GitHub Actions, Unit Testing, Interviewing, Amazon CloudWatch, Cloud9, AWS Command Line Interface (CLI), AWS Serverless Application Model (SAM), AWS Cloud Computing Services
GPU Software Consultant (via Toptal)
2023 - 2024
Blue Edge Financial LLC
- Leveraged OpenCL C and GPU parallel programming to decrease the time the client's MetaTrader expert advisors (EA) take to backtest trading strategies fourfold.
- Reviewed and profiled the single-thread baseline EA used by the client to spot its performance bottlenecks and design the new OpenCL module.
- Implemented the OpenCL C module as a combination of high-level OpenCL kernels for the parallel computing of input signals and the management of trades and low-level atomic operations implemented from scratch as self-contained C submodules.
- Integrated the OpenCL module to the client's EA by creating appropriate structures for data transfer between CPU and GPU and automatically handling possible errors.
- Debugged the OpenCL module and executed comprehensive tests to match the intermediate and final backtesting results between the baseline EA and its GPU-based version.
- Documented the new OpenCL module and trained the client's team on its use and maintenance.
Technologies: OpenCL, OpenCL/GPU, MQL5, NVIDIA CUDA, C++, Backtesting Trading Strategies, MetaTrader, MetaTrader 5, MQL, Software Development, Software Engineering, Parallel Programming, Low-latency Software, Low Latency, Debugging, Stock Trading, Trading, Stock Market, Stock Price Analysis, Clean Code, C
Senior Machine Learning Consultant (via Toptal)
2023 - 2024
Daphne Herlihy
- Implemented a proof of concept web app from scratch that generates personalized motivational audio tracks for patients going through therapy given their self-description and their therapist's prognostic.
- Assisted the client in elaborating the base template for the output audio script, including the definition of paragraphs, their goals, length, tone, and some positive and negative samples.
- Helped the client define the criteria used to evaluate the content of each audio track, such as their level of compassion, storytelling, coherence, and reliability, among others.
- Broke down the content generation task into subtasks to be solved individually via large language models, e.g., generating the opening section to ease the listener into the experience, the conclusion to send them off with a sense of achievement, etc.
- Applied prompt-engineering best practices, along with custom retrieval-augmented generation (RAG) and web crawling, to maximize the quality of the final audio content generated by the language models along each subtask previously defined.
- Kept optimizing and testing new approaches for prompt engineering and RAG by sharing results and getting the respective feedback from the client daily.
- Wrapped the audio generation pipeline as a Streamilit web application and ensured its modularity by following the clean code and architecture best practices.
- Optimized the UI and UX of the Streamlit web app to optimize the user's experience and ensure they would be able to provide the most pertinent feedback.
- Estimated the costs involved with each step executed by the final proof of concept when generating a new audio track; provided a full report to the client that discriminates each cost and a full description of the last system.
Technologies: Machine Learning, Artificial Intelligence (AI), Audio, Generative Pre-trained Transformers (GPT), OpenAI GPT-3 API, OpenAI GPT-4 API, Python 3, Python, Linux, Git, Software Development, Software Engineering, Software QA, Debugging, Profiling, Natural Language Processing (NLP), Data Cleaning, Data Collection, Web Scraping, Web Crawlers, Google APIs, Google API, Google Custom Search, Retrieval-augmented Generation (RAG), Prompt Engineering, Deep Learning, Deep Neural Networks (DNNs), Model Validation, Proof of Concept (POC), Web Development, Web App Development, Web App Deployment, App UI, OpenAI, Clean Code, Clean Architecture, Object-oriented Programming (OOP), AI Research, Project Management, Google Cloud Platform (GCP), System Design, AI Chatbots
Senior Machine Learning Consultant (via Toptal)
2023 - 2023
American Woodmark
- Implemented the end-to-end-inference code for a custom YOLOv8 object detection and mask segmentation model in Java to be deployed into a mobile robotics platform.
- Converted the model provided by the client to ONNX and implemented a baseline inference code in Python to confirm that the converted model's output matched that provided by the client.
- Set up the Java inference code as a Java project using Gradle and handled its major dependencies, including the ONNX runtime and OpenCV.
- Applied the model loading and inference in Java and the automatic check of their input and output tensor shapes.
- Implemented the model's output decoding, including the merge of detections and masks and their filtering according to the confidence threshold set by the user.
- Performed the decoding of output masks, which included computing their contours, getting their polygons via OpenCV, and filtering them via NMS.
- Plotted the model's output masks as binary and RGB image files and confirmed that all the intermediate and final outputs of the Java code match the reference Python baseline.
Technologies: Machine Learning, Python, Java, Android, OpenCV, Computer Vision, Computer Vision Algorithms, Image Processing, Robotics, Artificial Intelligence (AI), Gradle, Linux, Clean Code, Software Engineering, Software Development, Debugging, Project Management
Senior Machine Learning Consultant
2023 - 2023
Likken Norge
- Developed LykkenBot, an online multi-user virtual assistant POC that helps users at the Lykken Norge portal answer questions about higher education in Norway.
- Collected and formatted a test database of questions pertinent to the use case solved by LykkenBot.
- Automatically scraped web pages from target domains to access the knowledge required by the virtual assistant.
- Used OpenAI's embedding API and FAISS to create a vector store of scraped web pages to be queried by the virtual assistant during its execution.
- Used OpenAI's GPT API and LangChain to implement a conversational agent that powers the virtual assistant and plays the persona defined by the client.
- Engineered and refined the conversational agent's input prompt to optimize its responses' quality, length, and robustness.
- Used Python and Streamlit to implement and deploy the web app interface that runs the conversational agent under the hood.
- Used the validation dataset as the major reference for evaluating the final web app via end-to-end tests focused on the quality of the responses and overall application performance.
Technologies: Artificial Intelligence (AI), Machine Learning, Python 3, Python, Deep Learning, Natural Language Processing (NLP), Retrieval-augmented Generation (RAG), Chatbots, Chatbot Conversation Design, Prompt Engineering, OpenAI, OpenAI GPT-3 API, OpenAI GPT-4 API, ChatGPT, LangChain, Large Language Models (LLMs), Language Models, Web, Web Development, Web App Design, Data Cleaning, FAISS, Debugging, Software Development, Software Engineering, Proof of Concept (POC), Testing, Data Collection, Web Scraping, Web Crawlers, Data Scraping, Web App Deployment, Project Management, System Design, AI Chatbots, Website Data Scraping
Senior Machine Learning Consultant (via Toptal)
2023 - 2023
American Woodmark
- Developed an image segmentation model for a US-based enterprise kitchen and bath cabinet manufacturer to detect cabinet parts in their manufacturing facilities automatically.
- Provided the code for real-time object detection and segmentation inferences via YOLOv8 models converted to ONNX and TensorFlow Lite (TFLite) formats.
- Implemented the post-processing scripts to merge YOLO's bounding box and prototype mask outputs, convert those masks into probabilities, compute their polygonal arrays, and filter the final predictions via NMS.
- Compared the output masks of the converted models against the original Ultralytics PyTorch model to validate the conversion, inference, and post-processing scripts.
- Tuned some post-processing parameters to improve the quality of the output masks; documented and reviewed all the code.
Technologies: You Only Look Once (YOLO), PyTorch, Open Neural Network Exchange (ONNX), Python, TensorFlow, Machine Learning, Computer Vision, Image Processing, NumPy, Jupyter Notebook, GitHub, Git, Python 3, Real-time Computing, Object Detection, OpenCV, PIL, Matplotlib, Testing, Debugging, Computer Vision Algorithms, Project Management
Senior Machine Learning Consultant
2023 - 2023
Boris Rubinstein (Columbia University Irving Medical Center)
- Trained deep learning models for cross-subject emotion recognition via electroencephalography (EEG) classification.
- Reproduced the latest research papers on deep contrastive learning of subject-invariant EEG representations.
- Implemented a modular PyTorch framework to quickly experiment with new model architectures and train hyper-parameters when inducing cross-subject EEG classifiers.
- Designed and implemented the data loader, sampler, backbone model, and training modules, including their respective unit tests.
- Automated the exploration, formatting, and preprocessing of the EEG dataset provided by the client.
- Tracked the training and validation performances of EEG classifiers to fine-tune training hyper-parameters and obtain the best compromise between model bias and variance.
- Set up Git pre-commit hooks to automate the execution of unit tests and code formatting before updating the codebase with new commits.
- Demonstrated the delivered framework's use for training and evaluating EEG classification models to the client.
- Testimonial: "Michel picked up the details of a paper from an unfamiliar domain very quickly and implemented it flawlessly. It was a pleasure to work with him!" – Oliver Shetler, Senior Research Data Scientist, Columbia U. Irving Medical Ctr.
Technologies: PyTorch, Python 3, Pandas, NumPy, Matplotlib, Pytest, Deep Learning, Convolutional Neural Networks (CNNs), Data Preparation, Data Preprocessing, Python, Artificial Intelligence (AI), Artificial Neural Networks (ANN), Deep Neural Networks (DNNs), Model Validation, Benchmarking, EEG, Adam Optimization Algorithm, Signal Processing, Object-oriented Programming (OOP), Software Engineering, Software Development, Freelancing, Client Interviews, Client Interaction, Bug Fixes, Time Series, Time Series Analysis, Poetry, Unit Testing, Software Design Patterns, Conda, PIP, Data Science, Emotion Recognition, Classification Algorithms, Google Colaboratory (Colab), Digital Signal Processing, Planning, Classification, Software System Architecture Development, Software QA, Data Pipelines, Clean Code, Data Cleaning, Machine Learning, Pattern Recognition, Neural Networks, Optimization, Git, DevOps, Software Architecture, Linear Algebra, Presentations, DSP, Reporting, Multiprocessing, Signal Filtering, Supervised Machine Learning, Hyperparameters, Metric Learning, Data Validation, Data Visualization, Supervised Learning, GitHub, Documentation, Paper Prototyping, Windows Subsystem for Linux (WSL), Literature Review, Datasets, Algorithms, Data Structures, Frameworks, Desktop App Design, Automation, Testing, Debugging, AI Research, Technical Research, Feature Analysis, Project Management, System Design
Senior Machine Learning Engineer
2022 - 2023
SiDi (Samsung R&D Institute)
- Implemented speech processing systems based on cutting-edge AI research, including Samsung's HQ, to improve Bixby—Samsung's official voice assistant globally available on Galaxy devices.
- Reproduced, benchmarked, and improved state-of-the-art deep learning algorithms for supervised, contrastive, few-shot, and meta-learning tasks in the context of on-device speech processing.
- Collected, explored, preprocessed, and augmented public and in-house large-scale audio datasets.
- Implemented scalable pipelines to quickly explore different time-frequency feature representations for speech and signal transformations for data augmentation.
- Authored a new patent on automatic dataset creation for machine learning, which Samsung has filed in the United States Patent and Trademark Office (USPTO).
- Wrote highly modular and scalable code for training, regularizing, evaluating, and monitoring deep learning models in real-time audio streaming processing.
- Planned the technical roadmap, regularly wrote technical reports, and conducted weekly meetings with Samsung's HQ as the local focal point of contact.
- Assigned tasks and supervised a team of 10 engineers across different seniority levels.
- Authored research papers and presented them at relevant national and international speech and signal processing conferences.
- Delivered several internal and external technical lectures on machine learning and software development.
Technologies: Python, TensorFlow, NumPy, Pandas, Jupyter Notebook, Linux, Signal Processing, Audio Processing, Real-time Audio Processing, Scripting, Benchmarking, Speech Recognition, Python 3, Deep Learning, Deep Neural Networks (DNNs), Artificial Intelligence (AI), Edge Computing, DSP, Model Validation, UML, UML Diagrams, Git, Kanban, Jira, Kubeflow, High-performance Computing (HPC), Low-latency Software, Algorithms, Data Structures, Presentations, Reporting, Team Management, Interviewing, Research, Patents, Parallel Programming, Multiprocessing, Data Preparation, Data Preprocessing, Signal Filtering, Machine Learning, Supervised Machine Learning, Hyperparameters, TensorBoard, Metric Learning, Clustering, Data Validation, Data Visualization, Keras, Supervised Learning, Real-time Computing, Real-time Systems, Real-time Streaming, Real-time Data, Fourier Analysis, GitHub, GitOps, Team Mentoring, Software Engineering, Software Development, Agile Software Development, Object-oriented Programming (OOP), Prototyping, Paper Prototyping, Low Latency, Data Cleaning, Clean Code, Clean Architecture, Kubernetes, Docker, Windows Subsystem for Linux (WSL), Bixby, Data Pipelines, Time Series, Bug Fixes, Literature Review, Time Series Analysis, Audio, Conda, PIP, Poetry, Software Design Patterns, Multithreading, Proof of Concept (POC), Documentation, Mentorship, Data Science, Embedded Software, Embedded Systems, Real-time Embedded Systems, ARM Embedded, Embedded C, Slide, Embedded Linux, ARM Linux, PuTTY, SSH, Android Material Design, App UI, Classification Algorithms, Android UI Design, Android App Design, Android Development, Training, Graphics Processing Unit (GPU), Google Colaboratory (Colab), Digital Signal Processing, Profiling, Quantization, Planning, Video Editing, Classification, Debian, Statistical Methods, Statistical Significance, Software System Architecture Development, Software QA, Keyword Spotting (KWS), Cloud, Cloud Computing, User Feedback, Dashboards, Datasets, Labeling, Hypothesis Testing, XML, MVP Design, Model View Presenter (MVP), MVC Design, Mobile Development, Mobile App Development, Mobile App Design, Mobile, Android, Pattern Recognition, Neural Networks, GPU Computing, Optimization, FFmpeg, Matplotlib, Java, JUnit, Convolutional Neural Networks (CNNs), C, Software Architecture, NVIDIA CUDA, Librosa, CSS, HTML, Material Design, User Interface (UI), User Experience (UX), Vectorization, Linear Algebra, Artificial Neural Networks (ANN), Adam Optimization Algorithm, Google Material Design, Functional Programming, Debian Linux, Frameworks, Custom Audio Embedding, Data Scraping, Minimum Viable Product (MVP), Text-to-Speech (TTS), Speech Synthesis, Automation, Video Production, Testing, Data Collection, Debugging, AI Research, Technical Research, Feature Analysis, Data Engineering, Team Leadership, Project Management, Machine Learning Operations (MLOps), Speech-to-Text (STT), Data Analysis, System Design
Machine Learning Engineer
2020 - 2022
SiDi (Samsung R&D Institute)
- Engaged in global projects managed by Samsung, implementing and optimizing machine learning models for real-time speech recognition and audio event detection on mobile devices.
- Improved Galaxy's multilingual speech recognition features globally shipped by Samsung and currently used by millions of users.
- Applied weights quantization, pruning, and clustering to compress, optimize, and deploy multiple deep-learning models on edge devices for real-time speech processing.
- Implemented a modular proof of concept (POC) of Android applications for continually reading the device's microphone, feeding keyword-spotting models with the incoming audio streaming, and displaying its output predictions to the user.
- Employed Google Material Design to develop beautiful and user-friendly POCs of new keyword-spotting mobile applications for Samsung.
- Ensured the POC mobile applications followed clean architecture and implemented unit tests with more than 90% coverage on their domain layer.
- Assisted the Q&A team during live and simulated tests by specifying the clean and noisy scenarios, supervising its work, and collecting feedback.
- Documented the projects via text reports and UML diagrams frequently shared with Samsung's HQ.
- Used CircleCI to implement CI/CD pipelines and accelerate the validation of new features and updates on Samsung's codebase.
Technologies: Android, Java, Python, Python 3, UML, UML Diagrams, Jira, Git, JUnit, TensorFlow Lite, Signal Processing, Audio Processing, Real-time Audio Processing, Deep Learning, Speech Recognition, Artificial Intelligence (AI), Machine Learning, Prototyping, Proof of Concept (POC), Edge Computing, Android NDK, NDK, C, Google Material Design, Material Design, User Interface (UI), User Experience (UX), Object-oriented Programming (OOP), Software Engineering, Software Development, Agile Software Development, Kanban, Reporting, Documentation, Team Mentoring, Low Latency, Mobile, Mobile App Design, Mobile App Development, Mobile Development, Clean Code, Clean Architecture, Bug Fixes, Audio, Unit Testing, Software Design Patterns, Model View Presenter (MVP), MVC Design, Multithreading, Mentorship, CI/CD Pipelines, Continuous Integration (CI), CircleCI, ARM Linux, Embedded Software, Android Material Design, App UI, Classification Algorithms, Android UI Design, Android App Design, Android Development, Training, Digital Signal Processing, Memory Profiling, Profiling, Neural Network Pruning, Quantization, Planning, Classification, Software System Architecture Development, Software QA, Keyword Spotting (KWS), User Feedback, Data Science, Conda, Time Series Analysis, Literature Review, Time Series, Data Pipelines, XML, MVP Design, TensorFlow, Algorithms, Data Structures, Software Deployment, Pattern Recognition, Neural Networks, GPU Computing, Optimization, Convolutional Neural Networks (CNNs), Deep Neural Networks (DNNs), Software Architecture, Benchmarking, Research, Vectorization, Linear Algebra, Presentations, Data Preparation, Data Preprocessing, Artificial Neural Networks (ANN), Model Validation, DSP, Low-latency Software, Team Management, Multiprocessing, Supervised Machine Learning, Clustering, Data Validation, Data Visualization, Supervised Learning, Real-time Computing, Real-time Systems, Real-time Streaming, Real-time Data, GitHub, Paper Prototyping, Data Cleaning, Bixby, Frameworks, SSH, PuTTY, Embedded C, Embedded Systems, Real-time Embedded Systems, Embedded Linux, ARM Embedded, Slide, Minimum Viable Product (MVP), Automation, Testing, Debugging, AI Research, Technical Research, Feature Analysis, Project Management, Speech-to-Text (STT), Data Analysis, System Design, AI Chatbots
Teaching Assistant, Hardware/Software Interface
2019 - 2019
Universidade Federal de Sergipe
- Instructed around 30 computer science and computer engineering senior-year undergrads on the fundamentals of parallel and heterogeneous computing with OpenCL.
- Delivered lectures on the fundamentals of parallel computing, the architecture of OpenCL platforms, OpenCL execution and memory models, general optimization concepts, and performance benchmarking.
- Studied OpenCL's official documentation, including several online OpenCL repositories and articles, to develop the course material.
- Used LaTeX to elaborate clean and well-formatted slide presentations used as support material during the lectures.
- Implemented a public repository in C and C++ to demonstrate the concepts presented during the course.
- Elaborated coding practice exercises to be assigned to the students at the end of each lecture.
- Evaluated the students via regular practical exercises and a final project on parallel image processing via OpenCL.
- Dedicated weekly timeslots to assist the students with their assignments.
Technologies: C, C++, OpenCL, GPU Computing, Linux, Image Processing, Vectorization, Parallel Programming, Linear Algebra, University Teaching, Low Latency, LaTeX, Presentations, Assignments, Benchmarking, Mentorship, Slide, Training, Graphics Processing Unit (GPU), Google Slides, Digital Signal Processing, Open-source Software (OSS), Profiling, Planning, Bug Fixes, Multithreading, Clean Code, Data Structures, Signal Processing, Git, Data Preprocessing, Data Preparation, DSP, Low-latency Software, Reporting, Multiprocessing, Signal Filtering, Data Visualization, Real-time Computing, Real-time Systems, GitHub, Software Engineering, Software Development, Documentation, Team Mentoring, Algorithms, Testing, Debugging, Pointer Arithmetic, OpenCL/GPU, Project Management
Freelance Machine Learning Engineer
2017 - 2019
Viação Modelo
- Created a visual passenger counting system for a leading public transportation company operating the commute of half the population of Aracaju, Brazil.
- Researched new multiple-object tracking (MOT) algorithms based on cutting-edge computer vision and deep learning techniques.
- Developed SmartSORT, a new real-time tracking-by-detection algorithm based on deep-learning object detectors and hard-crafted spatial and temporal features.
- Collected hundreds of hours of real video recordings provided by the client from the internal surveillance system installed in their bus fleet.
- Annotated video recordings provided by the client and addressed people detection and tracking tasks using Labelbox and ViTBAT, respectively.
- Led and trained a team of three professionals to label and review the video recordings.
- Trained deep-learning visual object detectors, such as the R-CNN, YOLO, and SSD families, with the thousands of annotated frames of real passengers obtained from the client's video recordings using Amazon EC2 instances.
- Used the annotated videos of real passengers and the public MOT Challenge benchmark videos to train and evaluate SmartSORT, the real-time tracking algorithm developed during this engagement.
- Authored a research paper on the developed tracking algorithm, published in Springer Nature's Journal of Real-Time Image Processing.
- Used OpenCV, Qt, and Python to implement a POC system based on the developed tracking algorithm to automatically count bus passengers and export reports and dashboards on their flow throughout the day, given input video recordings.
Technologies: Computer Vision, Image Processing, Deep Learning, Object Detection, Object Tracking, Convolutional Neural Networks (CNNs), Deep Neural Networks (DNNs), Python, Python 3, NumPy, Pandas, Scikit-learn, OpenCV, Linux, Artificial Intelligence (AI), Team Mentoring, Low Latency, Edge Computing, Qt, Desktop App Development, Prototyping, Paper Prototyping, Client Interviews, Machine Learning, Object-oriented Programming (OOP), Software Engineering, Software Development, LaTeX, Benchmarking, Data Preparation, Data Preprocessing, Data Cleaning, Video Processing, Video Analysis, Image Analysis, Bug Fixes, Hypothesis Testing, Research, Labeling, Literature Review, Scripting, Proof of Concept (POC), Data Science, Video Streaming, Embedded Systems, Real-time Embedded Systems, Embedded C, Embedded Linux, ARM Linux, ARM Embedded, Embedded Software, PuTTY, SSH, FTP, Breadboarding, Optical Sensors, App UI, Classification Algorithms, Training, Graphics Processing Unit (GPU), Google Slides, Digital Signal Processing, Planning, Video Editing, Classification, Debian Linux, Debian, Arduino RTC, Statistical Methods, Statistical Significance, Software System Architecture Development, Software QA, Cloud, Cloud Computing, Dashboards, Seaborn, PIP, Conda, Time Series Analysis, Time Series, Data Pipelines, Remote Sensing, Clean Architecture, Clean Code, MVP Design, Model View Presenter (MVP), MVC Design, TensorFlow, PyTorch, Amazon Web Services (AWS), Algorithms, Data Structures, Software Deployment, Signal Processing, Pattern Recognition, Neural Networks, GPU Computing, Optimization, FFmpeg, Matplotlib, UML, UML Diagrams, Git, R, Software Architecture, Software Design Patterns, NVIDIA CUDA, Raspberry Pi, Parallel Programming, Full-stack, Full-stack Development, User Interface (UI), User Experience (UX), Amazon EC2, Linear Algebra, Presentations, Internet of Things (IoT), Sensor Data, Artificial Neural Networks (ANN), Model Validation, Adam Optimization Algorithm, DSP, High-performance Computing (HPC), Low-latency Software, Reporting, Team Management, Multiprocessing, Signal Filtering, Supervised Machine Learning, Hyperparameters, Metric Learning, Data Validation, Data Visualization, Supervised Learning, Real-time Computing, Real-time Systems, Real-time Streaming, Real-time Data, GitHub, Documentation, Freelancing, Client Interaction, Mentorship, Datasets, User Feedback, Image Recognition, Frameworks, Slide, Flask, Flask-RESTful, Image Compression, Desktop App Design, Minimum Viable Product (MVP), Automation, Video Production, You Only Look Once (YOLO), Testing, Data Collection, Debugging, Web App Design, Computer Vision Algorithms, AI Research, Technical Research, Tracking, Feature Analysis, Team Leadership, Project Management, Data Analysis, System Design, AWS Cloud Computing Services
Co-founder | Technical Leader | Full-stack Developer
2015 - 2018
Hype
- Co-founded Hype, a software development firm, to provide mobile and web development services.
- Developed a mobile social network for sharing and promoting local events. Ensured the app covered the main use cases: CRUD of users, events, post publications, and post comments.
- Employed Google Material Design and the hexagonal architecture to implement a use-case-driven modular Android app for the social network.
- Implemented a Java REST API to interface the Android app with the social network dataset implemented with PostgreSQL and hosted on AWS.
- Executed a web crawler with Java to automatically find new events promoted on famous websites and register them in the social network to solve the "chicken-egg" problem.
- Built a single-page application (SPA) to streamline a law firm's operations. Ensured the system covered CRUD operations on clients and legal processes. Used Angular in the front end and Laravel in the back end.
- Enforced a web crawler with Node.js to automatically monitor the status of the law firm's legal processes on the web pages of local legal courts, given the unavailability of official APIs.
- Used Bootstrap to implement a landing page for the law firm, which was integrated into the SPA and allowed clients to query the latest status of their legal processes.
- Led the entire technical development process of all projects executed by Hype, including the rise of requirements, use-case modeling, software architecture design, UI/UX, task assignments, and code review.
Technologies: Java, Android, Full-stack, Full-stack Development, PostgreSQL, REST, SQL, Databases, Material Design, User Interface (UI), User Experience (UX), Trello, Kanban, Unit Testing, Mockito, JUnit, Node.js, Amazon Web Services (AWS), Amazon EC2, Amazon S3 (AWS S3), Firebase, Web Crawlers, Team Management, Team Mentoring, Object-oriented Programming (OOP), Software Engineering, Software Development, Agile Software Development, MVC Design, Model View Presenter (MVP), MVP Design, Clean Code, Clean Architecture, XML, Google Material Design, Multithreading, Bug Fixes, Software Design Patterns, Mobile, Mobile App Design, Mobile App Development, Mobile Development, Documentation, Mentorship, PHP, Laravel, APIs, Web, Web Scraping, Web Services, MySQL, Relational Databases, Bootstrap, HTML, JavaScript, Angular, CSS, API Development, API Design, Reactive Programming, Single-page Applications (SPAs), Software QA, Software System Architecture Development, Slide, ARM Linux, PuTTY, SSH, FTP, Android Material Design, App UI, Android UI Design, Android App Design, Android Development, Training, Google Slides, Digital Production, Leads, Web Marketing, Planning, Web Server Development, Web App Development, Web Development, Cloud, Cloud Computing, User Feedback, Datasets, Functional Reactive Programming, Functional Programming, TypeScript, Linux, Data Structures, Software Deployment, UML, UML Diagrams, Git, Software Architecture, Scripting, Presentations, GitHub, Client Interviews, Client Interaction, Algorithms, Digital Product Design, Android Camera App, Amazon EC2 API, HTTPS, User Management, Data Scraping, Minimum Viable Product (MVP), Automation, Video Production, Testing, Debugging, Web App Design, Gradle, Web App Deployment, Data Engineering, Team Leadership, Project Management, Angular 2, System Design, Website Data Scraping, AWS Command Line Interface (CLI), AWS Cloud Computing Services
Computer Engineer (Intern)
2016 - 2016
BK Telecom
- Prototyped remote-sense devices for water supply and data center companies.
- Studied datasheets and manipulated digital and analogical sensors for measuring temperature, humidity, and water levels.
- Designed electronic circuits to serve as the interface between sensors and single-board computers, including the definition of appropriated resistors according to the point of operation required by all the components.
- Simulated the circuit interface using breadboards and validated the correctness of its theoretical design by measuring the components' point of operation and running end-to-end tests (i.e., reading the sensors via Raspberry Pi boards).
- Used Eagle PCB design and thermal transfer to build printed circuit boards for the circuit interface previously prototyped via breadboards.
- Welded all the required electronic components on the PCBs, including sensors, resistors, voltage regulators, and pin headers. Validated the welding of all the components and all the PCB routes by running end-to-end tests with Raspberry Pi boards.
- Set up new Raspberry Pi boards to continuously read the sensors welded on the PCBs and send the data to Zabbix servers in real time.
- Installed Debian operating system on new Raspberry Pi boards, set up SSH and FTP communication, integrated RTC modules, and implemented Python and Shell scripts to read sensor data using Raspberry's GPIO interface.
- Built prototype cases to accommodate Raspberry Pi boards and their respective PCB sensor interface. Installed and tested one of the prototypes in Universidade Federal de Sergipe to measure the temperature and humility of its data center in real time.
- Documented all the activities executed during the project via text reports and wrote comprehensive tutorials on the development of the prototypes for newcomers and other co-workers at the lab.
Technologies: Python, Bash Script, Shell, PCB Design, EAGLE, Electronics, Raspberry Pi, Internet of Things (IoT), Sensor Data, Research, Prototyping, Edge Computing, Remote Sensing, Arduino RTC, Debian, Debian Linux, Embedded Software, Embedded Systems, Real-time Embedded Systems, Embedded Linux, ARM Embedded, ARM Linux, PuTTY, SSH, FTP, Breadboarding, Digital Signal Processing, Software QA, PIP, Conda, Time Series Analysis, Bug Fixes, Time Series, Linux, Software Deployment, Git, Kanban, Scripting, Trello, Low Latency, LaTeX, Low-latency Software, Reporting, Signal Filtering, Real-time Computing, Real-time Systems, Real-time Streaming, Real-time Data, GitHub, Proof of Concept (POC), Software Engineering, Software Development, Documentation, Algorithms, Minimum Viable Product (MVP), Automation, Testing, Debugging
Undergraduate Researcher
2014 - 2016
Universidade Federal de Sergipe
- Explored the application of meta-heuristics to induce classification rules in the context of data mining.
- Modeled rules learning as a multi-objective discrete optimization problem and reproduced classical meta-heuristics, such as Random Search and Hill Climbing, to serve as baselines.
- Implemented a comprehensive C-based framework for loading structured and annotated datasets, learning new rules via meta-heuristics, and evaluating their quality when used to solve classification tasks.
- Implemented in C a new variation of particle swarm optimization (PSO) with multiple swarms for multi-objective discrete optimization based on Pareto optimality theory.
- Used GPU parallel computing to run and manage hundreds of swarms asynchronously via native CUDA C.
- Applied statistical hypothesis tests to benchmarked rules-based classifiers with different datasets.
- Wrote technical reports on the project methodology, planning, achievements, and results.
Technologies: Data Mining, Machine Learning, Pattern Recognition, Artificial Intelligence (AI), C, NVIDIA CUDA, GPU Computing, Weka, R, Research, Presentations, Explainable Artificial Intelligence (XAI), Association Rule Learning, Data Preparation, Data Preprocessing, Discrete Optimization, Metaheuristics, Evolutionary Computation, Evolutionary Algorithms, Benchmarking, Hypothesis Testing, Model Validation, Supervised Learning, Supervised Machine Learning, Parallel Programming, Optimization, Bug Fixes, Literature Review, Scripting, Data Science, Genetic Algorithms, Statistical Significance, Statistical Methods, Slide, Classification Algorithms, Graphics Processing Unit (GPU), Open-source Software (OSS), Memory Profiling, Profiling, Planning, Classification, Software QA, Datasets, Data Pipelines, Multithreading, Data Cleaning, Algorithms, Data Structures, Git, C++, Vectorization, Low-latency Software, Reporting, Multiprocessing, Hyperparameters, Data Validation, Data Visualization, GitHub, Software Engineering, Software Development, Documentation, Paper Prototyping, Desktop App Design, Testing, Debugging, Pointer Arithmetic, AI Research, Technical Research, Feature Analysis, Data Analysis
Experience
LykkenBot
LykkenBot is an online multi-user virtual assistant POC that helps users at the Lykken Norge portal to answer questions about higher education in Norway. It works based on a Q&A AI agent powered by a large language model (LLM) whose domain knowledge is retrieved from a preset vector store via retrieval augmentation generation (RAG). That POC was deployed as a web application that supports multiple sessions. LykkenBot POC significantly contributed to the client being selected for an acceleration program at Charge, a prestigious business development institution in Oslo, Norway.
Samsung Bixby's Custom Wake-up Phrase
Bixby is the voice assistant that helps users to interact via voice with their Samsung Galaxy devices. Custom Wake-up Phrase is a new feature launched in 2023 that enables users to customize the voice command spoken to activate Bixby instead of the usual "Hi Bixby." This feature allows on-device and fast wake-up phrase customization. Once enabled, it recognizes the user's custom command in real-time with high accuracy.
It was a privilege to be part of this project as a senior machine learning engineer and the focal point person at SiDi, Brazil. I worked on reproducing relevant AI and speech-processing research papers and implementing modular and scalable code bases for quickly training and evaluating new ML models. I also developed efficient pipelines for data loading, formatting, and augmentation and conducted brainstorms and technical discussions with the responsible Samsung team in Korea.
It was a privilege to be part of this project as a senior machine learning engineer and the focal point person at SiDi, Brazil. I worked on reproducing relevant AI and speech-processing research papers and implementing modular and scalable code bases for quickly training and evaluating new ML models. I also developed efficient pipelines for data loading, formatting, and augmentation and conducted brainstorms and technical discussions with the responsible Samsung team in Korea.
Samsung Galaxy's Camera Voice Control Upgrade
https://techwelkin.com/camera-voice-control-enable-samsung-galaxyGalaxy's camera voice control feature allows users to trigger Samsung's camera via voice; the feature is available in 14 languages.
As a machine learning engineer, I recently worked on upgrading this feature at SiDi, Brazil. My responsibilities included optimizing and compressing small-footprint deep learning models for keyword spotting. I led the development of a clean, modular, and scalable POC Android application, implemented its unit tests, and documented it for Samsung using UML diagrams and text reports.
As a machine learning engineer, I recently worked on upgrading this feature at SiDi, Brazil. My responsibilities included optimizing and compressing small-footprint deep learning models for keyword spotting. I led the development of a clean, modular, and scalable POC Android application, implemented its unit tests, and documented it for Samsung using UML diagrams and text reports.
SiDi KWS | Keyword Spotting Dataset
The world's largest dataset for keyword spotting, with 24+ million labeled spoken words across four languages.
I proposed and led the development of this project along with a team of engineers. I set up Montreal Forced Aligner (MFA), implemented and ran KeywordMiner—an open-source Python system that uses MFA to align speech files with their respective transcripts and takes that info to export isolated words—and wrote the research paper presented at Interspeech 2022.
I proposed and led the development of this project along with a team of engineers. I set up Montreal Forced Aligner (MFA), implemented and ran KeywordMiner—an open-source Python system that uses MFA to align speech files with their respective transcripts and takes that info to export isolated words—and wrote the research paper presented at Interspeech 2022.
AutoMedia | Automatic Video Creation
https://youtu.be/H4TUAlCQTdUAutoMedia is a Python application that automatically generates videos based on script output by ChatGPT. Its current implementation considers a "raking-video" template composed of an introduction section, followed by multiple ranking sections and a conclusion.
Given that template and the target video subject input by the user, AutoMedia employs ChatGPT and Google Images API to get the video script and the background images for each section, respectively. Moreover, it also uses an open-source TTS Python library to generate the speech included in the final video. Given that data (i.e., text script, images, and address), AutoMedia uses the MoviePy library to build the final video. As a plus, it also adds a pre-set background song. Its concept is similar to https://pictory.ai/
Given that template and the target video subject input by the user, AutoMedia employs ChatGPT and Google Images API to get the video script and the background images for each section, respectively. Moreover, it also uses an open-source TTS Python library to generate the speech included in the final video. Given that data (i.e., text script, images, and address), AutoMedia uses the MoviePy library to build the final video. As a plus, it also adds a pre-set background song. Its concept is similar to https://pictory.ai/
iModel | Virtual Try-on and Outlook Projection
https://youtu.be/7TZUyunnafsiModel is a Web-based SaaS that allows users to generate realistic images of themselves on new outlooks based on target pictures of professional models. Its goal is to help users save precious time when deciding on buying beauty services. It runs a state-of-the-art generative AI model on a custom API exclusively designed for this project. I have developed iModel from scratch as a side project and introduced it to some beauty salons in Brazil.
To build iModel, I have used React, Next.js, and Chakra UI in the front-end. Its back-end corresponds to a Python API implemented with Flask and NGINX, which runs on an AWS EC2 instance. That API is responsible for receiving a selfie and target images input by the user, feeding a local generative AI model, and returning the resulting image to the client app. I have also used Firebase to handle user authentication and management, as well as data analytics and the hosting of the client app.
To build iModel, I have used React, Next.js, and Chakra UI in the front-end. Its back-end corresponds to a Python API implemented with Flask and NGINX, which runs on an AWS EC2 instance. That API is responsible for receiving a selfie and target images input by the user, feeding a local generative AI model, and returning the resulting image to the client app. I have also used Firebase to handle user authentication and management, as well as data analytics and the hosting of the client app.
Automatic Bus Passengers Counting System
https://youtu.be/LxdM4vGCgjs/A computer-vision system for counting bus passengers based on the video recordings obtained from vehicle cameras. Its goal is to automate fraud detection since the client—a leading public transportation company in Aracaju, Brazil—had employed a whole department to manually monitor and register such occurrences daily.
As the responsible engineer, I collected real-world recordings from the client fleet, labeled the data, and trained ML models to detect the passengers on each video frame and associate those objects across frames. This engagement resulted in a research paper on a new tracking algorithm, published in the Journal of Real-Time Image Processing by Springer Nature.
As the responsible engineer, I collected real-world recordings from the client fleet, labeled the data, and trained ML models to detect the passengers on each video frame and associate those objects across frames. This engagement resulted in a research paper on a new tracking algorithm, published in the Journal of Real-Time Image Processing by Springer Nature.
TensorFlow Lite—The Professional Course
https://tflite-tpc.kpages.online/nova-pagina-1a945327-1475-472b-a912-1c1e3e2c4542/Because of the large size of the most successful machine learning models, AI-based mobile and embedding applications have relied on models stored on large remote servers. This brings many disadvantages to such applications, including the need for an internet connection, the latency caused by communication to the server, battery draining due to network connection, and sharing of user data.
To help engineers and scientists overcome these limitations, I co-authored an online course that demonstrates—with theoretical and practical examples—how to apply the most effective compression and optimization techniques to deploy state-of-the-art ML models 100% at the edge. The course goes beyond entry-level topics usually covered by most online tutorials on TensorFlow Lite. This course presents real Jupyter Notebooks and Android applications to illustrate the entire pipeline of model embedding and profiling used by industry professionals.
To help engineers and scientists overcome these limitations, I co-authored an online course that demonstrates—with theoretical and practical examples—how to apply the most effective compression and optimization techniques to deploy state-of-the-art ML models 100% at the edge. The course goes beyond entry-level topics usually covered by most online tutorials on TensorFlow Lite. This course presents real Jupyter Notebooks and Android applications to illustrate the entire pipeline of model embedding and profiling used by industry professionals.
Low-cost Autonomous Navigation System Based on Optical Flow Classification
https://youtu.be/hzyKAGhQExg/My final year undergraduate project involved developing an autonomous navigation system based on real-time obstacle detection via optical flow classification. This system was embedded into a real mobile robot controlled by a Raspberry Pi board.
During the first stage of this project, I built the vehicle using a robotics development kit for Arduino, along with a Raspberry Pi board fed by a portable power bank and connected to a USB camera. In the second stage, I used that vehicle to record POV videos while navigating in a lab with different obstacles. I used those videos to create a training dataset for an SVM classifier.
That classifier was fed with the magnitude and phase of optical-flow data points extracted from each frame in the video via the Lucas-Kanade algorithm implementation available in OpenCV. After training that model, I deployed it on the vehicle and evaluated its performance in avoiding new obstacles in real time while navigating in a lab.
This project resulted in the publication of a research paper and was named the Best Undergraduate Project Award at ERBASE 2017 by the Brazilian Computer Society in 2017.
During the first stage of this project, I built the vehicle using a robotics development kit for Arduino, along with a Raspberry Pi board fed by a portable power bank and connected to a USB camera. In the second stage, I used that vehicle to record POV videos while navigating in a lab with different obstacles. I used those videos to create a training dataset for an SVM classifier.
That classifier was fed with the magnitude and phase of optical-flow data points extracted from each frame in the video via the Lucas-Kanade algorithm implementation available in OpenCV. After training that model, I deployed it on the vehicle and evaluated its performance in avoiding new obstacles in real time while navigating in a lab.
This project resulted in the publication of a research paper and was named the Best Undergraduate Project Award at ERBASE 2017 by the Brazilian Computer Society in 2017.
On-device Image Classification via TensorFlow Lite
https://github.com/BytelandTechnologies/tflite-the-professional-course/A public repository I created as supplemental material for TensorFlow Lite—The Professional Course. Its goal is to demonstrate the use of TensorFlow Lite for real-time deep-learning model inference on edge devices using Android.
This repository already contains two fully functional Android applications that classify images of dogs and flowers offline and in real-time. Both apps load pre-trained deep-learning models already converted into TFLite files. The execution flow of both apps covers the loading of an input image via the device's camera or file storage system, its preprocessing, its classification via deep-learning model inference, the post-processing of the prediction scores outputted by the model, and the UI update to show the classification result.
This repository already contains two fully functional Android applications that classify images of dogs and flowers offline and in real-time. Both apps load pre-trained deep-learning models already converted into TFLite files. The execution flow of both apps covers the loading of an input image via the device's camera or file storage system, its preprocessing, its classification via deep-learning model inference, the post-processing of the prediction scores outputted by the model, and the UI update to show the classification result.
KeywordMiner Framework
https://github.com/michel-meneses/keyword-miner/A Python framework for automatically segmenting single-spoken words from transcribed speech recordings via automatic forced alignment. Its modular design allows support for different formats of input datasets.
I led the development of this framework to generate SiDi KWS, the world's largest dataset of single-spoken words, with 24+ million samples across four languages. My responsibilities included modeling the framework via object-oriented programming and UML diagrams, specifying the interface between its modules, implementing its classes, and documenting its repository.
I led the development of this framework to generate SiDi KWS, the world's largest dataset of single-spoken words, with 24+ million samples across four languages. My responsibilities included modeling the framework via object-oriented programming and UML diagrams, specifying the interface between its modules, implementing its classes, and documenting its repository.
Great OpenCL Examples
https://github.com/michel-meneses/great-opencl-examples/A GitHub repository that provides free, organized, ready-to-compile, and well-documented OpenCL C++ code examples.
OpenCL, or Open Computing Language, is a royalty-free framework for parallel programming of heterogeneous systems with different processing units, such as CPU, GPU, FPGA, and DSP. This repository serves as a reference for everyone interested in learning how to use OpenCL C++ to develop portable applications based on parallel computing. It was first created as a supplemental reference for the Hardware/Software Interface course offered by the computer science department of Universidade Federal de Sergipe during the first semester of 2019.
OpenCL, or Open Computing Language, is a royalty-free framework for parallel programming of heterogeneous systems with different processing units, such as CPU, GPU, FPGA, and DSP. This repository serves as a reference for everyone interested in learning how to use OpenCL C++ to develop portable applications based on parallel computing. It was first created as a supplemental reference for the Hardware/Software Interface course offered by the computer science department of Universidade Federal de Sergipe during the first semester of 2019.
Education
2017 - 2019
Master's Degree in Computer Science
The Federal University of Sergipe - Sergipe, Brazil
2012 - 2017
Bachelor's Degree in Computer Engineering
The Federal University of Sergipe - Sergipe, Brazil
Certifications
JANUARY 2024 - PRESENT
Advanced C Programming: Optimize Performance and Efficiency
SEPTEMBER 2023 - PRESENT
LangChain 101 for Beginners (OpenAI / ChatGPT / LLMOps)
Udemy
APRIL 2023 - PRESENT
Create an Open-Source Project in Python
APRIL 2023 - PRESENT
Testing Python Data Science Code
SEPTEMBER 2022 - PRESENT
Software Development Life Cycle (SDLC)
SEPTEMBER 2021 - PRESENT
Advanced Predictive Modeling: Mastering Ensembles and Metamodeling
AUGUST 2021 - PRESENT
Advanced NLP with Python for Machine Learning
AUGUST 2021 - PRESENT
Algorithmic Trading and Finance Models with Python, R, and Stata Essential Training
JULY 2021 - PRESENT
Building Deep Learning Applications with Keras 2.0
MARCH 2021 - PRESENT
Building and Deploying Deep Learning Applications with TensorFlow
OCTOBER 2020 - PRESENT
Device-based Models with TensorFlow Lite
DeepLearning.AI
SEPTEMBER 2020 - PRESENT
Programming Foundations: Algorithms
SEPTEMBER 2020 - PRESENT
Programming Foundations: Data Structures
SEPTEMBER 2020 - PRESENT
Programming Foundations: Design Patterns
SEPTEMBER 2020 - PRESENT
Software Architecture: Patterns for Developers
MARCH 2020 - PRESENT
DevOps Essentials
4linux
Skills
Libraries/APIs
TensorFlow, NumPy, Scikit-learn, Matplotlib, Keras, API Development, Amazon EC2 API, PyTorch, Pandas, OpenCV, Arduino RTC, MoviePy, Google API, Google APIs, FFmpeg, Node.js, NDK, Flask-RESTful, React, PIL, HubSpot API, vLLM
Tools
Git, Trello, LaTeX, TensorBoard, GitHub, Google Slides, Slide, Jira, Pytest, Shell, Seaborn, PuTTY, Firebase Authentication, You Only Look Once (YOLO), Weka, EAGLE, Android NDK, CircleCI, ChatGPT, NGINX, Open Neural Network Exchange (ONNX), Gradle, Notion, AWS Command Line Interface (CLI), Amazon SageMaker, Amazon CloudWatch, GraphRAG, AWS CloudFormation
Languages
Python 3, Python, Java, UML, C, SQL, JavaScript, CSS, HTML, Bash Script, XML, Embedded C, R, C++, PHP, TypeScript, Octave, MQL5, MQL
Frameworks
JUnit, TensorFlow Lite, Mockito, AWS Serverless Application Model (SAM), Angular, Bootstrap, Flask, Angular 2, OpenCL, Qt, Laravel, Next.js, Serverless Framework, LangGraph, Streamlit
Paradigms
Unit Testing, Kanban, Agile Software Development, Object-oriented Programming (OOP), REST, Real-time Systems, Mobile App Design, Mobile Development, MVC Design, Model View Presenter (MVP), Clean Code, Clean Architecture, Android Material Design, Automation, Testing, Scrum, Parallel Programming, High-performance Computing (HPC), DevOps, Desktop App Development, Functional Programming, Functional Reactive Programming, Continuous Integration (CI), Reactive Programming, Web App Design, Serverless Architecture, Continuous Deployment
Platforms
Android, Jupyter Notebook, Software Design Patterns, Amazon EC2, Mobile, AWS Cloud Computing Services, Linux, Amazon Web Services (AWS), NVIDIA CUDA, Firebase, Web, Debian, Debian Linux, ARM Linux, Embedded Linux, Google Cloud Platform (GCP), Heroku, Raspberry Pi, Kubeflow, Kubernetes, Docker, MetaTrader, MetaTrader 5, Azure, AWS Lambda, Weights & Biases
Storage
PostgreSQL, Data Validation, Data Pipelines, Relational Databases, Databases, Amazon S3 (AWS S3), LVM, MySQL
Industry Expertise
Project Management
Other
Deep Learning, Computer Vision, Object Tracking, Algorithms, Data Structures, Software Deployment, Signal Processing, Image Processing, Machine Learning, Artificial Intelligence (AI), Pattern Recognition, Neural Networks, Optimization, UML Diagrams, Object Detection, Convolutional Neural Networks (CNNs), Deep Neural Networks (DNNs), Software Architecture, Scripting, Benchmarking, Research, Material Design, Vectorization, Linear Algebra, Low Latency, Presentations, Data Preparation, Data Preprocessing, Artificial Neural Networks (ANN), Model Validation, Adam Optimization Algorithm, Edge Computing, Low-latency Software, Reporting, Team Management, Multiprocessing, Supervised Machine Learning, Hyperparameters, Metric Learning, Clustering, Data Visualization, Supervised Learning, Real-time Computing, Real-time Streaming, Prototyping, Proof of Concept (POC), Google Material Design, Software Engineering, Software Development, Documentation, Team Mentoring, Paper Prototyping, Mentorship, Data Cleaning, Video Processing, Video Analysis, Image Analysis, Mobile App Development, MVP Design, Association Rule Learning, Discrete Optimization, Metaheuristics, Evolutionary Computation, Evolutionary Algorithms, Bug Fixes, Labeling, Literature Review, Audio, Conda, PIP, Data Science, Datasets, Keyword Spotting (KWS), APIs, API Design, Software QA, Software System Architecture Development, Classification, Planning, Quantization, Neural Network Pruning, Profiling, Digital Signal Processing, Graphics Processing Unit (GPU), Training, Android Development, Android App Design, Android UI Design, Classification Algorithms, App UI, Breadboarding, Image Recognition, Frameworks, Custom Audio Embedding, HTTPS, User Management, Data Scraping, Minimum Viable Product (MVP), Debugging, Data Collection, Pointer Arithmetic, AI Research, Technical Research, Tracking, Feature Analysis, Data Engineering, Team Leadership, Data Analysis, System Design, AI Chatbots, Website Data Scraping, Electronics, GPU Computing, Speech Recognition, Trading, Natural Language Processing (NLP), Audio Processing, Poetry, Data Mining, Real-time Audio Processing, Robotics, Full-stack, Full-stack Development, User Interface (UI), User Experience (UX), University Teaching, Assignments, Internet of Things (IoT), DSP, Interviewing, Patents, Signal Filtering, Real-time Data, Fourier Analysis, Freelancing, Client Interviews, Client Interaction, Multithreading, Explainable Artificial Intelligence (XAI), Hypothesis Testing, Windows Subsystem for Linux (WSL), Bixby, Time Series, Time Series Analysis, Dashboards, User Feedback, Cloud Computing, Cloud, Web Development, Web App Development, Web Server Development, Video Streaming, Web Scraping, Web Services, Single-page Applications (SPAs), Genetic Algorithms, Statistical Significance, Statistical Methods, Video Editing, Educational Videos, Online Course Design, Memory Profiling, Digital Product Design, Digital Production, Google Colaboratory (Colab), Android Camera App, Optical Sensors, DC Motor Drive, FTP, SSH, Embedded Software, Embedded Systems, Real-time Embedded Systems, ARM Embedded, Firebase Hosting, Image Compression, Desktop App Design, Text-to-Speech (TTS), Speech Synthesis, Video Production, Knowledge Bases, Computer Vision Algorithms, Web App Deployment, OpenCL/GPU, Machine Learning Operations (MLOps), Speech-to-Text (STT), Finance, Ensemble Methods, CI/CD Pipelines, Support Vector Machines (SVM), Librosa, Web Crawlers, PCB Design, Sensor Data, EEG, GitOps, Remote Sensing, Emotion Recognition, Trailer, Web Marketing, Leads, Open-source Software (OSS), OpenAI GPT-4 API, OpenAI GPT-3 API, LangChain, Chatbots, Q&A Bots, Generative Pre-trained Transformers (GPT), Image Generation, SaaS, SaaS Design, Chakra UI, Deepfake, Amazon API Gateway, Amazon Route 53, SSL Certificates, Language Models, Generative Artificial Intelligence (GenAI), Retrieval-augmented Generation (RAG), Chatbot Conversation Design, Prompt Engineering, OpenAI, Large Language Models (LLMs), FAISS, Google Custom Search, Backtesting Trading Strategies, Stock Trading, Stock Market, Stock Price Analysis, Serverless, GitHub Actions, Pinecone, AI Website Builders, Multimodal Models, A/B Testing, Exploratory Data Analysis, Cloud9, R&D, AI Testing, FastAPI, Serverless GPUs, SGLang, Multimodal GenAI, Fine-tuning, Open-source LLMs, Data Labeling, Transformers, AI Agents, Agentic AI, Vector Databases, Weave, Intercom, RAG Systems, RAG Pipelines, RAG Architecture, WandB, Embedding Models
Collaboration That Works
How to Work with Toptal
Toptal matches you directly with global industry experts from our network in hours—not weeks or months.
1
Share your needs
Discuss your requirements and refine your scope in a call with a Toptal domain expert.
2
Choose your talent
Get a short list of expertly matched talent within 24 hours to review, interview, and choose from.
3
Start your risk-free talent trial
Work with your chosen talent on a trial basis for up to two weeks. Pay only if you decide to hire them.
Top talent is in high demand.
Start hiring