I’m an AI/ML engineer who builds production systems, not prototypes. From an enterprise legal platform indexing 50,000+ French-language documents to real-time retrieval pipelines, my work is about making LLMs reliable once they leave the notebook.
My edge is owning the full stack: I architect the retrieval pipeline, tune the prompts, build the FastAPI backend, and ship it in Docker. No handoffs, no gaps between “it works in the demo” and “it works in production.”
Building production RAG systems, AI agents, and LLM pipelines for clients worldwide. Top Rated, 100% Job Success across 23+ delivered engagements.
Led AI product development: demand-forecasting models and a multi-endpoint conversational AI system serving production traffic.
Specialized in AI, data science, and algorithms. IBM-certified data scientist.
I work across the whole retrieval pipeline — from embeddings and vector search to the agents and backend that make it production-ready.
Enterprise legal AI platform
A full-stack RAG system processing 50,000+ French legal documents: triple-index Elasticsearch search, three-tier RBAC, Claude-powered anonymization, OHADA compliance, and ~23 documented endpoints.
Agentic orchestration on LangGraph
An agent platform built on LangGraph's StateGraph with a Model Context Protocol client for external tool access, async SQLite-backed checkpointing for persistent multi-session memory, and human-in-the-loop interrupts before sensitive tool calls execute.
Natural-language cloud automation
A chat-driven AWS assistant on Amazon Bedrock (Claude 3.5 Sonnet) with LangChain memory, letting non-technical users create, update, and manage AWS resources through natural language instead of the CLI, with mandatory field confirmation before any resource-changing action executes.
Recovery-support SaaS backend
Backend for a 24/7 AI sponsor chatbot for 12-step recovery programs: Claude-based chat with prompt caching, Stripe subscription billing with webhooks, and a WhatsApp voice pipeline combining Whisper transcription with synthesized voice replies.
Earnings-call intelligence with CrewAI
A CrewAI multi-agent pipeline that ingests earnings-call transcripts and financial PDFs, delegates summarization and structuring across specialized agents, and outputs client-ready report documents.
AI therapy chatbot (MVP)
A mental-health support chatbot with mood assessment and tracking, CBT/mindfulness-based guidance, session memory, and automatic risk detection that surfaces crisis helplines. Shipped production-ready with documented performance targets.
CNN diagnostics for healthcare
CNN-based classification models for brain-tumor detection and diabetic retinopathy screening from medical scans, built with TensorFlow using transfer learning.
Scanned documents, made retrievable
A five-stage pipeline (PDF, image, preprocessing, OCR, structuring) converting scanned documents into clean, vector-store-ready JSON for downstream RAG ingestion.
Human-in-the-loop LinkedIn & job-application agent
A local-first, three-agent tool that drafts LinkedIn posts/comments and tailored job-application cover letters, with web-search-grounded topic scoring. Every publish/apply action stays manual, avoiding platform ToS violations.
Conversational voice support
A voice support agent that answers inbound calls, handles FAQ and order-lookup requests through natural conversation, and escalates to a live agent when a request falls outside its scope.
Lessons from shipping a 24/7 WhatsApp voice pipeline — Whisper transcription, Claude-generated replies, and the prompt-caching setup that cut input-token costs by roughly 90%.
How I designed an agent platform on LangGraph's StateGraph with Model Context Protocol tool access, persistent checkpointed memory, and human approval steps before sensitive actions.
A practical breakdown of when RAG outperforms fine-tuning and vice versa, based on real production experience building both types of systems.
How I architected a production RAG system processing 50,000+ French legal documents using Claude Sonnet and Pinecone with 95%+ retrieval accuracy.
Available for freelance work and select full-time opportunities. Typical response time is within 24 hours.