Most ML projects die in notebooks. Mine don't. I sit at the intersection of machine learning engineering and product thinking — taking data from raw to reliable, whether that means a RAG pipeline answering questions over complex documents, a ML model predicting health risk, fraud, or churn, or an LLM evaluation framework catching reasoning failures before they reach users. I build things that actually work outside the lab.

"A model that can't be deployed is just a very expensive experiment."

• 🏗️ Building production-grade ML pipelines with observability, evaluation, and monitoring baked in
• 🧠 Exploring LLM reasoning diagnostics — understanding why models fail, not just when
• 🔍 Developing RAG systems that go beyond naive retrieval — semantic chunking, hybrid search, real-time ingestion
• 📊 Automating ML quality gates to catch prompt regressions before they ship

End-to-end ML system · 88.5% accuracy · FastAPI + Streamlit + SHAP + MLflow + Docker

A production-ready healthcare ML system built the right way — not just a notebook, but a deployable service with explainability (SHAP), experiment tracking (MLflow), a REST API, an interactive dashboard, and full Docker support. Built for trust, not just performance.

Research-grade diagnostics for multi-step reasoning failures in LLMs

Most LLM evals tell you a model is wrong. This framework tells you where the reasoning broke down. Designed to surface failure modes in chain-of-thought reasoning — useful for anyone building reliable LLM-powered products.

Production RAG system · FastAPI + React + ChromaDB + LLM observability

A full-stack document Q&A system with integrated LLM observability and monitoring. Goes beyond basic retrieval with real observability into how the system answers — not just what it answers.

Automated prompt regression detection · CI/CD for LLMs

A production-grade evaluation pipeline that automatically catches quality regressions before prompts reach users. Think CI/CD, but for LLM behavior — because shipping a broken prompt is just as bad as shipping broken code.

Languages     │ Python · SQL
ML & AI       │ PyTorch · scikit-learn · Hugging Face · LLMs · RAG · AI Agents
MLOps         │ FastAPI · MLflow · Docker · GitHub Actions
Data & BI     │ Pandas · NumPy · Tableau · Amazon QuickSight
Cloud         │ AWS · GCP · BigQuery
Explainability│ SHAP

I'm open to collaboration on:

• Production ML systems — from model to API to deployment
• AI-powered data products — real-time insights, dashboards, automation
• LLM evaluation & reliability — making AI systems you can actually trust
• Real-world automation — turning workflows into intelligent pipelines

If you're building something ambitious and need an ML engineer who thinks beyond the notebook, let's talk.

📬 nwangumaemmanuel29@gmail.com