project / Dec 2025 – Jan 2026

Policy Radar

Full-stack federal policy research agent that searches nine official government sources and streams source-backed answers with full-text and PDF retrieval.

view github reporesume
timelineDec 2025 - Jan 2026
focusai, fullstack
stack
  • Python
  • FastAPI
  • TypeScript
  • React
  • Vite
  • Tailwind CSS
  • SQLite
  • ChromaDB

problem

Federal policy activity is fragmented across official sources, making it slow to track rulemakings, bills, spending, datasets, DOJ releases, and related documents in one research workflow.

approach

A FastAPI tool-execution layer routes questions across nine official U.S. government sources, while a React client streams answers, citations, and live tool activity and preserves research sessions for follow-up work.

implementation notes

  • Built a FastAPI backend with REST and SSE chat endpoints for sessions, messages, streaming responses, source storage, model configuration, and content fetching.
  • Integrated official-source tools for Regulations.gov, GovInfo, Congress.gov, Federal Register, USAspending, Treasury Fiscal Data, data.gov, DOJ, and Search.gov.
  • Supported OpenAI Responses/Chat Completions, Anthropic, Gemini, and OpenAI-compatible custom endpoints through configurable provider settings.
  • Extracted full text from web and PDF sources, then indexed PDF chunks with ChromaDB for session-scoped retrieval.
  • Protected document retrieval with SSRF checks and domain allowlisting, and added authentication and rate limiting around the research API.
  • Built a React/Vite/TypeScript interface with chat history, editable messages, source cards, a live work log, provider/model settings, and source/time filters.

impact

  • Unified research workflow across regulation, legislation, spending, fiscal-data, dataset, and enforcement sources
  • Inspectable citations, extracted source text, persistent histories, and session-scoped PDF retrieval

links