Software & AI Engineer
I build production software end-to-end across product, infrastructure, and applied AI. Eight years engineering systems that move from architecture to shipped product.
View Work →ABOUT me
Software engineer with 8 years shipping production systems across healthcare AI, biotech, and enterprise. My work spans backend APIs, React frontends, LLM-integrated pipelines, and the cloud infrastructure underneath — taking complex systems end-to-end.
Most recently at PictorLabs.ai— a UCLA spinoff in digital pathology — fine-tuning Google's PathFoundation vision model, building LLM tooling with Claude API and LangGraph, and leading delivery of an AI virtual staining pipeline for whole slide images.
I'm at my best where AI engineering meets real production engineering — training pipelines, retrieval systems, API design, and the infrastructure that holds it all up. Engineering fundamentals first, frameworks second.
Production systems end-to-end
AI fundamentals from first principles
Product, infra, and model pipelines
WORK history
SELECTED work
Codebase reasoning engine that indexes any GitHub repo and answers natural-language questions with source citations and line numbers. Every retrieval layer hand-built — no LangChain, no LlamaIndex. tree-sitter AST chunking, contextual LLM descriptions (35–49% retrieval improvement per Anthropic benchmarks), Qdrant hybrid search (dense + BM25 + RRF), Cohere cross-encoder reranking, HyDE + query expansion, and a full ReAct agent via MCP with 10 tools, working memory, and parallel async execution.
AI virtual HE staining pipeline for unstained brightfield whole slide images, built at PictorLabs.ai (UCLA spinoff, venture-backed). Owned end-to-end: fine-tuned Google's PathFoundation vision model in PyTorch, built Django API with organ/species/diagnosis-based predictor routing, Kafka job orchestration, and TorchServe model serving.
Closed-source · proprietaryRAG system for querying research papers — grounding answers in document content with full source citations. Custom semantic search (MiniLM embeddings + ChromaDB), BM25 keyword search, and hybrid retrieval via RRF fusion. No LangChain. Precursor to Cartographer — built to understand retrieval internals from first principles.
04Full encoder-decoder transformer implemented in PyTorch — multi-head attention, positional encoding, layer norm, greedy decoding, label smoothing, LR warmup. Trained EN→ES on the Opus Books dataset, end-to-end from "Attention Is All You Need".
05Q-learning, DQN (Breakout), and PPO (LunarLander) trained on Gymnasium with Stable Baselines 3 — plus a custom multi-agent dual-taxi environment built from scratch (observation space, reward shaping, training loop). Domain breadth: not just LLMs.
TECH stack
build
TOGETHER
If you're building production AI systems and want someone who can go from model training to shipped product — reach out.