Ata S.

I build production RAG systems and LLM agents that get deployed, evaluated, and maintained, not just demoed. Most recently I spent 20 months as an AI engineer on a legal-tech SaaS for Italian law firms, building an agentic RAG workflow that reviews legal documents against Italian legal sources, plus the evaluation framework (MLflow, LangSmith) and the AWS infrastructure (Terraform) behind it. WHAT I'VE SHIPPED • Legal document review: an agentic RAG workflow (LangGraph) that checks contracts against retrieved Italian legal sources, plus an AI legal assistant with streaming responses and conversation memory. • Product recommendations: a multi-agent LangChain workflow that interprets what a shopper wants and returns structured recommendations from Amazon marketplace data, served through a FastAPI backend. • Content generation: a DSPy pipeline that turns restaurant-owner data into complete multi-section websites, and a CrewAI agent team that writes Amazon product descriptions. • Data Q&A: RAG over CSV and Excel files, with code-generation chains for calculations and batch testing that tracks accuracy and response time. HOW I WORK • Evaluation from day one (MLflow, LangSmith), so you can see whether each change made the answers better or worse. • Infrastructure as code on AWS (Terraform, Docker, GitHub Actions), so what you get is deployable and repeatable, not a notebook. • 7+ years of backend work in Python (FastAPI, Django, Flask) and Node (Express.js), so the AI sits on a solid application. GOOD FIT FOR • A RAG chatbot or assistant over your documents • AI agents that automate a multi-step workflow • Fixing an LLM app that gives wrong or made-up answers • Taking a prototype to production on AWS TOOLS Python, LangChain, LangGraph, CrewAI, DSPy, OpenAI API, Anthropic API, AWS Bedrock, Pinecone, PostgreSQL, MongoDB, Redis, FastAPI, Express.js, Terraform, Docker, GitHub Actions