About
The LP Diligence Agent is a portfolio prototype demonstrating an agentic-AI workflow applied to private-markets diligence. Given a private equity fund quarterly report, the agent runs a nine-item diligence checklist and returns citation-backed structured answers with explicit confidence tags.
Why this exists
Built as a portfolio piece demonstrating an agentic-AI workflow over LP-format private-equity quarterly reports. The corpus uses publicly available materials: PSERS FOIA-released quarterly reports from 2017 plus SEC 10-Q filings from Blackstone Private Equity Strategies Fund (2025).
Architecture
- Python backend: FastAPI + sqlite-vec + sentence-transformers
- Multi-step agent: one retrieval + one Anthropic Claude call per checklist item
- MCP server: same tools exposed for Claude Desktop
- Eval harness: 20-question golden set scored by an LLM judge
- Guardrails: citation-required answers, refusal-by-default on missing data