Current Work

Building LLM agents and document-intelligence pipelines for research, legal, contracts, and healthcare workflows, alongside AI tooling for aviation operations — maintenance, scheduling, and regulatory document intelligence. My platforms ingest PDFs end-to-end — classification, clause extraction, and risk evaluation — backed by RAG with hybrid retrieval and reranking, structured citations, and RAGAS evaluation. A frontier/local routing layer pairs cloud models with on-prem open-weight models so regulated workloads stay private without sacrificing frontier-quality output.

Knowledge Base

This site is a working reference of notes, diagrams, and code patterns accumulated while designing and operating data and ML systems in production. It is organized by domain and kept close to the primary sources.

Area	Coverage
Flagship Project	End-to-end architecture deep dive on MyDocumentIntelligence.com: ingestion, hybrid retrieval + reranking, frontier/local LLM routing, structured citations, RAGAS evaluation, and AWS deployment.
LLM Engineering	Production patterns for 2026: agents and Model Context Protocol (MCP), function calling and structured output, RAG evaluation with RAGAS, hybrid search and reranking, vLLM and quantization, LangGraph and DSPy.
AI & Machine Learning	Retrieval-augmented generation, vector stores (FAISS, Chroma, pgvector), neural-network architectures, natural-language processing, recurrent models, Hugging Face tooling, and document-ingestion pipelines.