Hemen Ashodia
Verified Expert in Engineering
Full-stack Developer
Ahmedabad, Gujarat, India
Toptal member since April 3, 2018
Hemen has over 14 years of experience in full-stack development and 10 with machine learning, contributing to hundreds of ML projects. He possesses hard-to-find expertise in artificial neural networks, deep learning, reinforcement learning, and generative adversarial networks. Hemen has also worked for companies that were later acquired by Intel, mainly for their machine learning development. He has worked with notable companies like Bitcoin.com, Tala, and Johnson & Johnson.
Portfolio
Experience
- Python - 14 years
- Deep Neural Networks (DNNs) - 11 years
- C++ - 11 years
- Generative Artificial Intelligence (GenAI) - 10 years
- PyTorch - 8 years
- TensorFlow - 8 years
- OpenAI - 5 years
- Large Language Models (LLMs) - 4 years
Preferred Environment
Microservices, Django, React, Python, PyTorch, Retrieval-augmented Generation (RAG), OpenAI GPT-4 API, Llama 3, Qdrant, Pinecone
The most amazing...
...thing I've invented is the fastest data structure algorithm, h+Tree, which is 3x faster than b+Tree—saving 40% to 66% in energy/server cost.
Work Experience
Chief Scientist and Software Engineer
F(x) Data Labs Pvt., Ltd.
- Invented the fastest data retrieval algorithm, h+Tree, with an up to 300% speed increase compared with the conventional b+tree. This algorithm is useful for databases like MySQL, PostgreSQL, MongoDB, SQLite, and more. It was ported to C/C++ and Python.
- Created a full-stack public cloud on the latest version of OpenStack with infrastructure as a service, database as a service, and object storage service. The system was built with OpenStack, Python, Ansible Notebooks, Ubuntu 16.04, and unit tests.
- Developed an artificial intelligence-based enterprise resource planning (ERP) system for a 60-year-old real estate company with the help of TensorFlow, a deep neural network, Laravel, Angular 4, Frappe, Python, MySQL, Ubuntu 16.0, and unit tests.
- Built a loan prediction default system with an accuracy of 99.4% with the help of TensorFlow, Python, and R. It also predicted the loan default amount with a 96.3% accuracy.
- Implemented baby-cry detection with sound classification and support vector machines (SVM) on Raspberry Pi to notify parents when a baby is crying.
- Developed video classification with a deep neural network to detect different human gestures from the video as small clips; used an artificial neural network, Caffe 2.0, C/C++, a convolutional neural network, and NVIDIA CUDA.
- Created a hospital operation room scheduler with artificial neural networks, Python, Laravel, Angular 4, CSS3, Vagrant, and unit tests.
- Enabled video style transfers for 3D videos to an artificial neural network with PyTorch, Google Cloud, CUDA, and more. Implemented a movie review sentiment analysis with a convolutional neural network, TensorFlow, Python, and Python Notebooks.
- Created a 3D face mesh model from a single mobile camera-based selfie with more than 98% accuracy. The project was built using PyTorch, a convolutional neural network, and CUDA.
- Worked on multiple LLM-based projects, like a search engine based on ChatGPT and GPT-4. Used LangChain in the back end. Created an AI layer that uses privately available case data and indexes it efficiently to reduce the cost and integrates with GPT-4.
Staff Machine Learning Engineer
Tala
- Optimized the usage of a Snowflake data warehouse to reduce annual cost by $2 million by changing query patterns.
- Invited by Google to present our real-time inference infrastructure design.
- Added support for LangChain in the feature engine to do credit underwriting.
- Migrated a monolith to microservice infrastructure with Kubernetes. Led and coordinated with a team of 40 data scientists and MLEs.
- Sped up credit underwriting time from six seconds to less than one second. Sped up the deployment pipeline from a 4-week deployment cycle to 10 minutes.
- Created a feature engine framework that supported real-time inference of 30+ models in four countries for loan underwriting.
- Designed and built an end-to-end deployment pipeline for model management. Integrated Kubeflow and Metaflow. Built a feature logic framework to manage countries, models, and versions of the model.
Lead Engineer and Engineering Manager
Bitcoin.com
- Led a team of six full-stack developers, SDETs, and data scientists to create a completely new casino called bitspincasino.com. Added new games to an existing casino (games.bitcoin.com).
- Managed Bitcoin and Bitcoin Cash Nodes for a self-made payment gateway, saving on processing fees.
- Worked on a tech stack of microservices, such as Vue, Python, Django, and Spring Boot.
- Created AI-based slot games to manage better RTP games, as well as built an AI back end to predict fraud while gambling. Prevented frauds worth millions every year.
- Added new games like slots, blackjack, poker, Angry Banker, etc. based on PixiJS.
Lead Machine Learning Engineer
Braided Data
- Built a document processing system from the ground up to make a PDF document searchable with OCR and perform redaction in real time. Implemented an OCR system capable of processing millions of pages that had already processed several billion words.
- Trained several states of the art in text processing GPT-2 and BERT models to perform high-accuracy tag recognization. Invented a patentable real-time NER algorithm to process and learn without catastrophic interference.
- Customized the source code of the largest OCR tool in the market using Tesseract to allow table recognition. Built a Kubernetes cluster to perform on-demand autoscaling and processing.
- Led the team between multiple projects and coordinated a number of projects within different time zones.
Lead Data Scientist and Machine Learning Expert
Johnson & Johnson India
- Led and created data pipelines which used millions of data points for the identification of the best strategies to increase the effectiveness of professional educational events.
- Verified the work done by another data science company in the domain of medical professional education events.
- Worked directly with principal scientists of Johnson & Johnson's Data Science in New York, specifically on events for medical device training all across the USA and elsewhere.
Lead Go Developer
Loom Network
- Built a Karma system for a decentralized crypto platform.
- Implemented a Sparse Merkle Tree.
- Developed a rate limiter for access control.
- Designed with Oracle for a Loomchain platform.
Freelance TensorFlow Expert
Ever AI (via Toptal)
- Worked for Ever AI, the world's leading face recognition technology company that raised a total of 30 million in funding. It implements face recognition at a planetary scale of 12 billion images.
- Was tasked to invent the fastest algorithm to identify the correct face using machine learning.
- Developed the fastest algorithm in less than half the time with a 99.81% accuracy.
Chief Technology Officer
Zidisha, Inc.
- Built new features and functionalities for a website with Laravel, Propel, Bootstrap, Beanstalk servers, and Ubuntu/trusty64.
- Supported the entire back-end code and reviewed the front-end code that was done by other team members for further verification.
- Implemented automatic language translation for each new user.
- Worked as a remote engineer for a US-based company.
Founder and Lead Back-end Developer
Remarkin
- Developed a video engagement prediction system with artificial neural networks, OpenCV, and Scala.
- Constructed an interactive, gamification-based course for C/C++ with PHP, MN Framework, jQuery, JavaScript, and CSS3.
- Built a remote interactive C/C++ compiler with a TinyC compile and a GNU compiler.
Project Trainer
Aspire Institute
- Trained more than 150 students during a year for their final year projects in the college.
- Coached for C/C++ language programming skills.
Web Developer
Cruxbot
- Built a Parallax website that represented an artificial intelligence-based company.
- Helped build an interface for the text summarization browser extension of CruxLight.
Web Developer
Amitech
- Developed Picinchat.com and enabled it to create mega emoticons from users' uploaded images. These emoticons were created using PHP, the Facebook API, and the GD library.
- Created a Ridje browser extension to allow clients to customize any web page that they visit and keep changes/customization persistent. This was built using JavaScript and the Chome extension developer toolkit.
Freelance Developer
Self-employed
- Created a real-time chatting feature for a site with a real-time feed of a user typing on the other side.
- Developed a Flash-based cricket single player vs computer game.
- Worked on a tic-tac-toe human vs computer machine.
- Built a live video-streaming website with a screenshot for video scroll navigation.
Experience
Braided Data Solutions
F(x) Data Cloud
https://fxdatalabs.comInvented h+Tree
Zidisha, Inc.
https://zidisha.org/Cruxbot
BAKERI ERP | Artificial Intelligence-based ERP
Artudata
I built the dashboard for the company, which provides marketing and advertisement management and analytics with Google AdWords APIs and the Facebook app. It provides a fully automated marketing experience managed with machine learning.
Korebin
TensorFlow Compilation with a GPU for macOS X High Sierra
https://gist.github.com/Extrodox/c7dd4a146ba6175415b4b5f569b5c934As TensorFlow does not get compiled for the graphics processing unit (GPU) for Mac operating systems anymore, this file is a manual walkthrough I created for compiling TensorFlow for MacOS 10.13.
MN Framework
https://github.com/Extrodox/MN_FrameworkEver.ai | Face Recognition at a Planetary Scale
https://ever.ai/Loom Network
https://loomx.io/Education
Research Fellowship in Startup Innovation
Stanford University via Venture Studio - Ahmedabad, India
Bachelor of Engineering Degree in Information Technology
Lalbhai Dalpatbhai College of Engineering - Ahmedabad, India
Certifications
Leadership and Excellence Award
Gujarat Technological University, Ahmedabad, Gujarat, India
Skills
Libraries/APIs
PySpark, React, Node.js, REST APIs, WebGL, Keras, jQuery, Three.js, PhantomJS, Laravel Elixir, Cinder, Python API Bindings for OpenStack, OpenCV, NumPy, Natural Language Toolkit (NLTK), OpenNLP, TensorFlow, SQLAlchemy, FFmpeg, Parallax, Auth, GD Library, Pandas, Propel, Stripe, PyTorch, Stanford NLP, Vue, Vue 3, Scikit-learn, XGBoost, OpenAI Assistants API
Tools
RabbitMQ, Celery, BigQuery, Jira, Asana, Confluence, Apache, Composer, Shell, VirtualBox, MATLAB, Stanford CoreNLP, TensorBoard, Git, Gulp, Subversion (SVN), Mathematica, Scilab, Vagrant, GitHub, Flash, Microsoft Access, Microsoft Unit Tests, OpenAI Gym, Apache Airflow, DeepSeek
Languages
SQL, GraphQL, TypeScript, JavaScript, R, PHP, Python, SCSS, Less, HTML5, CSS3, C, C++, Snowflake, Solidity, Java, Go, Sass, Scala, HTML, CSS, Python 3, DYNAMO
Frameworks
Django, Apache Spark, Laravel, AngularJS, Angular, Bootstrap 3, Selenium, Nova, Caffe, Laravel 5, Next.js, Bootstrap, Tendermint, AWS HA, React Native, Flutter, Redux, PHP MVC, PixiJS, Metaflow, LlamaIndex
Paradigms
ETL, Agile Project Management, REST, Agile Software Development, Test-driven Development (TDD), Functional Programming, Imperative Programming, Design Thinking, Design Patterns, Gamification, Microservices, Human-computer Interaction (HCI), Unit Testing, Agile Product Management, Responsive
Platforms
Kubernetes, Amazon Web Services (AWS), Docker, Google Cloud Platform (GCP), WooCommerce, WordPress, LAMP, OpenStack, Arduino, Ubuntu, Azure, Windows Server, Ethereum, Jupyter Notebook, Red Hat Linux, NVIDIA CUDA, Kubeflow, Databricks
Storage
Redshift, MariaDB, MySQL, Redis, Oracle 10g, PostgreSQL, MySQL Clustering, Neo4j, Cloud Deployment, MongoDB, InfluxDB, LevelDB, Propel ORM, Amazon DynamoDB, MySQLi
Industry Expertise
Project Management
Other
Generative Pre-trained Transformer 2 (GPT-2), Custom BERT, Data Science, Data Engineering, Algorithms, Quantitative Finance, Algorithmic Trading, Artificial Intelligence (AI), Computer Vision, Machine Learning, Task Management, Team Management, CTO, Progressive Web Applications (PWAs), Product Management, Leadership, eCommerce, Natural Language Processing (NLP), Convolutional Neural Networks (CNNs), Macros, User Experience (UX), Networks, Azure Virtual Networks, Routing, Virtualization, Neutron, Ceilometers, HEAT, OpenStack Swift, Deep Neural Networks (DNNs), Generative Adversarial Networks (GANs), Capsule Networks, Adam Optimization Algorithm, Standardization, WordNet, Deep Learning, Generative Pre-trained Transformers (GPT), OpenAI GPT-4 API, Full-stack, Large Language Models (LLMs), Large Language Model Operations (LLMOps), OpenAI, Meta Llama, Retrieval-augmented Generation (RAG), CI/CD Pipelines, FastAPI, Containers, AI Chatbots, Chatbots, Generative Artificial Intelligence (GenAI), Neural Networks, Data Structures, Responsive Design, Artificial Neural Networks (ANN), Advance, Facebook, Chrome Extensions, Ajax, Raft Consensus Algorithm, Linear Regression, Logistic Regression, Data Modeling, Windows 10, Tesseract, Optical Character Recognition (OCR), Information Technology, Trusty64, PayPal, Software Development Lifecycle (SDLC), Language Models, Natural Language Understanding (NLU), Data Analysis, Graphics Processing Unit (GPU), Unsupervised Learning, Fraud Detection, Casino Games, 2D Games, 3D Games, LangChain, OpenAI GPT-3 API, System Architecture, Llama 3, Llama 2, BERT, Entrepreneurship, Mining, Clustering, K-means Clustering, ERPNext, Random Forests, Qdrant, Pinecone, Computer Science, Computer Engineering, Architecture, APIs
How to Work with Toptal
Toptal matches you directly with global industry experts from our network in hours—not weeks or months.
Share your needs
Choose your talent
Start your risk-free talent trial
Top talent is in high demand.
Start hiring