Michael G.

Hi, I'm Michael. I'm an AI engineer who builds LLM agents, RAG systems, and the evaluations/benchmarks that prove they work. For the past 18 months I was the lead developer for all things AI at AlphaScript, a US-based fintech focused on investment banking. I built agentic features on top of OpenAI-compatible APIs, designed the custom tooling that let the models take real actions in the product, built Airflow ETL pipelines for data processing and curation, and mentored and worked with other developers. Before AI, I spent years shipping full-stack TypeScript products, so I can build the whole thing, not just the prompt. What I build: - Agentic workflows: tool/function calling, custom agent loops, multi-agent setups, and MCP servers that connect models to your tools and data. I build directly on the OpenAI API spec when you need full control, and use LangGraph or LangChain when they're the right fit. I work with OpenAI, Claude and other OpenAI-compatible models. - RAG and LLM understanding: ingestion, chunking, embeddings and retrieval (pgvector or a dedicated vector database) that give grounded answers from your own documents and data. - LLM evaluation and benchmarks: I build Terminal-Bench, MLE-bench and multi-agent style tasks for evaluating AI coding agents, including sandboxed Docker environments, automated graders, reference solutions, and difficulty calibration against frontier models. - The product around it: APIs, dashboards and integrations in Python/FastAPI, Node.js/NestJS and React/Next.js, deployed on AWS. What my last client said: "Michael was our lead developer for all things AI at AlphaScript, and he was exceptional. He has an impressive ability to think through complex AI problems quickly, identify thoughtful solutions, and turn them into products of outstanding quality." How I work: - I start by defining what "good" looks like for you, then measure against it, so we know the AI is actually getting better. - Clear, regular updates and honest estimates. - I work with a small number of long-term clients so each one gets my full attention. Technologies: - AI: OpenAI and Claude APIs, function/tool calling, MCP, LangGraph, LangChain, RAG, embeddings, pgvector and vector databases, prompt engineering, LLM evals, machine learning - Data pipelines: Apache Airflow ETL, data processing and curation - Backend: Python, FastAPI, Node.js, NestJS, Express, PostgreSQL, MongoDB, Firebase - Frontend: TypeScript, React, Next.js, Vue/Nuxt, Tailwind CSS - Infra and quality: Docker, Kubernetes, AWS, GitHub Actions, unit and end-to-end testing For an AI feature that you want to build, or an agent you need to measure properly, send me a short note about what you're building and I'll tell you honestly how I'd approach it.