I build grounded, inspectable AI systems for knowledge, documents, and real-world workflows.
Capable AI is not enough. Production systems should be grounded in evidence, clear about uncertainty, and designed around human judgment.
I am a Computer Science PhD researcher working across retrieval-augmented generation, document intelligence, knowledge systems, and human-in-the-loop automation. My work connects research ideas with reliable software that people can inspect, evaluate, and use.
| Research direction | What I work on |
|---|---|
| Reliable RAG | Grounded retrieval, citation quality, evaluation, and evidence-aware refusal |
| Document intelligence | Extraction and validation pipelines for invoices, contracts, PDFs, and business records |
| Human-centered automation | Reviewable workflows that preserve control around consequential external actions |
| Applied AI engineering | FastAPI services, local-first systems, model integration, testing, and production-minded demos |
A cohesive portfolio of three independently runnable, local-first AI products. Each project includes an English interface, documented API, tests, sample data, Docker configuration, and a client-facing demo path.
| Product | Workflow | Trust boundary |
|---|---|---|
| InquiryPilot | Gmail inquiry → qualified lead → Sheets record → Gmail draft | Deduplication, explicit approval, draft-only automation |
| ExtractFlow | PDF, invoice, or contract → structured review queue | Source evidence, confidence scoring, validation, XLSX/CSV/JSON export |
| CiteDesk | Business documents → grounded multi-turn answers | File/page citations and low-evidence refusal |
Evidence before confidence · Human review before external action
Local-first when privacy matters · Clear handoff before hidden complexity
Languages & systems
Python · JavaScript · FastAPI · SQLite · Chroma · Docling · n8n · Docker
AI engineering
RAG · Embeddings · Information Extraction · Evaluation · Prompt Engineering · OpenAI API
I am open to research collaboration and selective engineering work involving RAG, document processing, internal knowledge systems, and reviewable AI automation.
Start with the working systems: code · product previews · architecture · demo guide
For a project conversation, open an issue with a short, non-confidential description of the workflow you want to improve. Please never include credentials, customer records, or private documents in a public issue.
Reliable AI systems · Designed for evidence, control, and real use