I architect end-to-end ML pipelines for high-stakes tabular data, deploy RAG knowledge systems, and fine-tune LLMs. Bridging the gap between raw data and production-ready intelligence.
Production-grade ML pipelines with a focus on robust evaluation, probability calibration, and AI agent orchestration.
End‑to‑end pipeline predicting heart disease from 13 clinical features. Focuses on high-stakes medical decision support via rigorous probability calibration. 15 model versions evaluated to find the optimal trade-off between Brier Score and F1.
A universal RAG (Retrieval-Augmented Generation) Knowledge Ingestion engine. Vacuums codebases, docs, and agent logs into a ChromaDB vector store. Features an LLM reporting tool (Llama-3.3) and a live CLI dashboard for monitoring knowledge saturation.
Built an asynchronous multi-agent system serving 4 specialized agents (Strategist, Executor, Guardian, Researcher). Fully integrated with the ChromaDB RAG to ensure "RAG Before Action" protocols and automated peer-review of ML pipelines.
Real-time 3-way comms between Antigravity IDE, CLI and Gemini CLI —
via work_log.md.
Real-time dynamic monitoring of AI static evaluation rules via Mistral-Large semantic Layer 3 audits. Live TrustScore decays penalize context blindness.
| Rule ID | Description & Intent | Metrics | Trust Score |
|---|---|---|---|
| Loading Ledger... | |||