Leonardo is available for hire
Hire LeonardoLeonardo dos Santos Pinheiro
Verified Expert in Engineering
Machine Learning Engineer and Software Developer
Brisbane, Queensland, Australia
Toptal member since July 4, 2019
Bio
Leonardo is a senior AI engineer with 10+ years of experience shipping production AI, including six years delivering engagements through Toptal. He specializes in LLM agents, retrieval-augmented generation (RAG), and LLM evaluation and LLMOps, working end-to-end from prototype to deployed services. At Microsoft, Leonardo built an LLM evaluation infrastructure. On recent Toptal engagements, he built RAG copilots and automated MLOps pipelines.
Portfolio
Next Intelligence GmbH
Artificial Intelligence (AI), AI Engineering, AI Agents...
Microsoft
TypeScript, Azure, PyTorch, ChatGPT, Natural Language Processing (NLP), OpenAI...
OLLA BOWLS - FZCO
Artificial Intelligence (AI), Python, Web Development, Facial Recognition...
Experience
- Python - 10 years
- Artificial Intelligence (AI) - 10 years
- Deep Learning - 8 years
- Natural Language Processing (NLP) - 4 years
- Large Language Model Operations (LLMOps) - 4 years
- Computer Vision - 3 years
- AI Agents - 3 years
- Reinforcement Learning from Human Feedback (RLHF) - 2 years
Preferred Environment
Windows Subsystem for Linux (WSL), Cursor AI, Codex, Claude Code
The most amazing...
...thing I've done is develop end-to-end machine learning pipelines for cancer detection—from data selection and labeling to deployment on Kubernetes.
Work Experience
AI Engineer/Architect
2026 - 2026
Next Intelligence GmbH
- Built a Python agent plane on AWS Bedrock AgentCore (EU) using LangGraph Deep Agents, with durable processing runs, per-run trace and cost persistence, LangSmith observability, and provider-swappable inference across Claude and OpenAI.
- Added human-in-the-loop agent actions covering email send, calendar appointments, Microsoft To Do, and document request emails, with edit-before-approve drafts and just-in-time OAuth scope grants.
- Redesigned the Ask assistant as a bounded RAG over a tenant transcript corpus, then extended it with multi-turn sessions and a stateful document agent.
Technologies: Artificial Intelligence (AI), AI Engineering, AI Agents, Large Language Models (LLMs), Technical Architecture, Retrieval-augmented Generation (RAG), Agentic RAG Systems, Scalability, Enterprise ready, Data Protection, Small Language Models (SLMs), Startups, Claude Code, Claude API, Claude, Recall, Transcription, General Data Protection Regulation (GDPR), Node.js, React, TypeScript, Electron
Senior AI/ML Engineer
2023 - 2026
Microsoft
- Helped move a flagship video-composition Copilot from fixed ML/LLM pipelines to an agent-based architecture—context management, subagents, and tool and skill design.
- Built LLM evaluation infrastructure (eval design, LLM-as-a-judge, and CI-gated quality) adopted across the product organization, replacing ad-hoc "vibe checks" with measurable iteration.
- Shipped LLM features for AI video composition: intent detection, commanding, Q&A, chaptering, and summarization over video transcripts.
- Worked as an open-source contributor to Microsoft AutoGen—multi-agent orchestration examples and LLM clients.
- Integrated on-device audio models (noise cancellation, audio classification, and voice-activity detection) via ONNX Runtime with Azure back-end services.
Technologies: TypeScript, Azure, PyTorch, ChatGPT, Natural Language Processing (NLP), OpenAI, AI Agents, Large Language Model Operations (LLMOps), Large Language Models (LLMs), Multi-agent Systems, Prompt Engineering, LangChain
Senior AI Developer
2024 - 2024
OLLA BOWLS - FZCO
- Developed web scrapers that scanned websites with potential content leaks for analysis.
- Built a large-scale scraping and automated analysis pipeline processing millions of link analyses per day for roughly 800 creators using NLP and face recognition analysis.
- Implemented a LangGraph workflow for website classification that identified leak candidates through in-context learning.
- Implemented a React web browsing agent to extract and ingest DMCA information for reporting using deep agents.
- Owned end-to-end system design—ingestion, data pipelines, APIs, and cloud infrastructure.
Technologies: Artificial Intelligence (AI), Python, Web Development, Facial Recognition, Scraping, LangGraph, Agentic AI, Large Language Models (LLMs), Prompt Engineering, LangChain, Workflow
AI-first Software Engineer
2024 - 2024
Trilogy
- Built chat, RAG-based Q&A, and summarization features for the Jive CoPilot assistant using the OpenAI API.
- Specified AI pipelines to automate software-engineering tasks—test generation, documentation generation, and PR descriptions.
- Triaged and fixed Copilot defects across Jive Cloud and Jive Hop.
Technologies: Java, OpenAI, AWS Serverless Application Model (SAM), AWS Lambda, JavaScript, RAG Pipelines, LLM Integration, Large Language Models (LLMs), Retrieval-augmented Generation (RAG), Prompt Engineering
NLP Expert via Toptal
2024 - 2024
Kwan Ting Chang
- Worked with the client to create a dataset for TTS. Helped with script generation, voice actor hiring, data distribution analysis, speech-and-text emotion analysis, and data preprocessing.
- Fine-tuned StyleTTS 2 model for text-to-speech with emphasis on capturing the diversity of emotional tones.
- Delivered a full TTS system for integration with a larger web application based on Python, PyTorch, FastAPI, and Docker.
Technologies: Artificial Intelligence (AI), Natural Language Processing (NLP), Speech-to-Text (STT), StyleTTS, Tortoise TTS, Tacotron 2
Senior ML Engineer
2021 - 2023
HARRISON-AI
- Handled the end-to-end pipeline for cancer detection, including labeling with V7, model building for semantic segmentation using Hugging Face, and deployment on Kubernetes using FastAPI.
- Contributed to an evaluation system for the CT brain classification model. Built internal Python library for multiple hypothesis statistical testing.
- Led a series of transformer-based experiments for embryo selection to improve the performance of a production system based on Inception3D. The ViViT experiments led to a better model, which was later moved to production.
- Built a chat application based on semantic search using an LLM to aid clinicians in finding medical reports containing specific cases of interest. The application was used to perform case selection for image labeling.
- Contributed to a chatbot for retrieving medical cases based on semantic search using OpenAI ChatGPT and LangChain.
Technologies: TensorFlow, PyTorch, Amazon Web Services (AWS), Docker, Artificial Intelligence (AI), Machine Learning, Chatbots
Senior ML Engineer
2021 - 2023
Jungle Scout
- Developed an MLOps system for automatic model evaluation and promotion for an eCommerce weekly model training pipeline using SageMaker, MLflow, Kedro, and Airflow.
- Productized deep learning models for time series eCommerce models based on PyTorch Lightning and SageMaker.
- Created a new model serving pipeline based on the MLflow model registry and SageMaker endpoints, with Lambda and an API gateway for scaling and Datadog monitoring.
- Created a model performance dashboard based on Plotly Dash and Redis. Deployed on Fargate.
Technologies: Amazon Web Services (AWS), TensorFlow, Docker, Deep Learning, Python, Machine Learning, Time Series
Senior Software Engineer (Data and AI)
2019 - 2021
BCGX
- Used stereo vision and image segmentation on satellite imagery to aid an infrastructure company in vegetation management. The system was used to map the risk of vegetation encroachment with assets.
- Developed a Twitter analysis dashboard to measure tweet sentiments, a network of influencers, and visualize trends per tag/time to aid strategic designers in research.
- Developed a gradient-boosting model for activity classification using sensor data for a supply chain startup. The system was used to track illegal activity at different points in the supply chain.
- Built image classification models for crop recognition and crop pest/disease recognition for a farming startup. The system supported advisory for smallholder farmers in Southeast Asia.
- Built a recommender system for a cashback program startup, enabling personalization of content to drive engagement in the platform.
- Created a performance dashboard for a farming startup using Data Studio and BigQuery.
Technologies: Amazon Web Services (AWS), Docker, TensorFlow, OpenCV, Python, Machine Learning, JavaScript, Node.js, Azure
Senior Machine Learning Consultant
2018 - 2019
Servian
- Developed and deployed a churn model using gradient boosting for an insurance company.
- Developed and deployed a convolutional network for customer spending forecasting using TensorFlow, Ansible, Docker, ECS, DynamoDB, and PostgreSQL.
- Developed and deployed a text classification system using a convolutional model using TensorFlow and Spark.
- Designed a data science strategy for a major financial institution. Mentored junior data scientists.
- Explored a large corpus of insurance claims data using association rule mining, topic modeling, semantic similarity, and other text mining techniques.
- Created a open domain chatbot based on machine comprehension (Facebook's DrQA) using PyTorch, Flask, React, and DialogFlow.
- Assisted in the development of a person tracking system using Yolo v2 and Kalman filters for a major Australian retail company.
- Assisted with a markdown system based on demand forecasting using Facebook's Prophet and revenue optimization using mixed-integer linear programming.
Technologies: Amazon Web Services (AWS), TensorFlow, Hadoop, Spark, Scala, Python, Agile Data Science, Machine Learning, Chatbots
Data Scientist
2017 - 2018
Mojo Power
- Developed and deployed a serverless linear model for load forecasting using Python, NumPy, and AWS Lambda.
- Created a proof-of-concept Hidden Markov Model for load disaggregation.
- Developed a model for credit scoring of energy customers.
- Developed and deployed an LSTM model for load forecasting using PyTorch.
- Developed dashboards for analytics reporting on energy usage using Tableau.
- Used topic modeling for exploratory data analysis of customer reviews.
- Worked on a PoC for solar panel detection on satellite images using Facebook's Detectron.
Technologies: PostgreSQL, AWS Lambda, Python, Data Science, Machine Learning
Quantitative Developer
2016 - 2017
Macquarie Bank
- Parsed and analyzed unstructured data of logs of order execution into SQL Server.
- Back-tested optimal execution strategies.
- Developed a Plotly dashboard to visualize market data.
- Tested and investigated new trading strategies.
- Tested machine learning algorithms for commodities trading.
Technologies: Plotly, Vagrant, Microsoft SQL Server, Python
Quantitative Researcher
2012 - 2016
Comissão de Valores Mobiliários
- Developed regulatory research studies using statistical modeling (estimation and hypothesis testing).
- Created market risk reports and visualizations with time series analysis and forecasting using R.
- Elaborated a risk monitoring system using Monte Carlo simulation and statistical estimation using Java.
- Developed a data warehouse to aggregate data related to market risk and development of BI reports using BusinessObjects.
- Led a data governance group to discover and catalog data sources across the whole organization.
Technologies: Microsoft 365, Python, Microsoft SQL Server, SQL
Business Analyst
2010 - 2012
Brazilian Institute of Metrology
- Worked with operational teams to develop scripts to collect metrics from operational processes and automate reporting. Scripts were based on Python and PowerShell.
- Built Cognos dashboards to monitor business KPIs related to operational processes. Also made custom Jupyter Notebooks for additional data analysis.
- Created a simplified database/warehouse using SQLite to aggregate data from multiple spreadsheets and support live business dashboards.
Technologies: Dashboards, Python, Windows PowerShell, SQL
Experience
Investment Funds Program
A program that creates a network of interconnections between investment funds and simulates a cascading failures algorithm. The program serves as a stress testing model that needs data to run and can be executed from the command line by typing "python fin_contagion_inv_funds.py."
Education
2014 - 2016
Master's Degree in Applied Math
Getulio Vargas Foundation - Rio de Janeiro, Brazil
2006 - 2009
Bachelor's Degree in Management Science
Getulio Vargas Foundation - Rio de Janeiro, Brazil
Certifications
DECEMBER 2018 - DECEMBER 2020
AWS Certified Developer
AWS
Skills
Libraries/APIs
Scikit-learn, XGBoost, Pandas, NumPy, SpaCy, OpenCV, TensorFlow, SciPy, Natural Language Toolkit (NLTK), Node.js, PyTorch, Flask-RESTful, Tortoise TTS, Claude API, React
Tools
Jupyter, Plotly, GitLab CI/CD, Git, Amazon Elastic Container Service (ECS), Apache Airflow, Vagrant, AWS CloudFormation, PyCharm, MATLAB, ChatGPT, Codex, Claude Code, Claude
Languages
Python, SQL, Scala, Bash, TypeScript, JavaScript, Java
Frameworks
Spark, Hadoop, Windows PowerShell, Flask, AWS Serverless Application Model (SAM), LangGraph, Electron
Paradigms
Scrum, Kanban
Platforms
Amazon Web Services (AWS), Docker, AWS Lambda, Linux, Google Cloud Platform (GCP), Apache Kafka, Visual Studio Code (VS Code), Amazon EC2, Azure
Storage
Amazon S3 (AWS S3), Amazon DynamoDB, InfluxDB, PostgreSQL, Microsoft SQL Server, MongoDB, Neo4j
Other
Data Science, Artificial Intelligence (AI), Dashboards, Agile Data Science, Machine Learning, Natural Language Processing (NLP), Computer Vision, Deep Learning, Generative Pre-trained Transformers (GPT), OpenAI, AI Agents, Large Language Model Operations (LLMOps), Large Language Models (LLMs), Multi-agent Systems, Prompt Engineering, LangChain, Workflow, A/B Testing, Visualization, Statistics, APIs, Scraping, Analytics, Dashboard Design, Chatbots, Reinforcement Learning from Human Feedback (RLHF), Retrieval-augmented Generation (RAG), Microsoft 365, Recommendation Systems, Windows Subsystem for Linux (WSL), Amazon RDS, ECS, Time Series, Optimization, Graphs, Speech-to-Text (STT), StyleTTS, Tacotron 2, Cursor AI, Web Development, Facial Recognition, Agentic AI, RAG Pipelines, LLM Integration, AI Engineering, Technical Architecture, Agentic RAG Systems, Scalability, Enterprise ready, Data Protection, Small Language Models (SLMs), Startups, Recall, Transcription, General Data Protection Regulation (GDPR)
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