FundFinder

Know who funds the work before you ask them to.

FundFinder researches a company's India CSR activity, scores its fit against your mission, and tells you who to talk to. It reads annual reports so you don't have to.

Scores are opinions with arithmetic. Every claim cites a source. Verify before outreach.

What it does

Mode	Purpose	Output
Screen	Fast yes/no on a prospect	Score, tier, summary
Deep Research	Full dossier	Score, evidence, decision-makers, .docx report, .xlsx workbook

How it works

company name
   -> fetch sources (CSR pages, MCA, annual reports, partners, people)
   -> extract facts (LLM)  -> targeted follow-up searches
   -> score against criteria (LLM + deterministic rules)
   -> cited report, saved to history

If evidence is thin, the app says so instead of guessing. Silence is not a negative signal.

Stack

FastAPI, Jinja2 + HTMX, Supabase (Postgres, Storage, Google OAuth), Anthropic API, Google Programmable Search.

Quick start

pip install -r requirements.txt
cp .env.example .env        # fill it in
uvicorn app.main:app --reload

Environment

Variable	Notes
SUPABASE_URL, SUPABASE_KEY	Server-side service key
SUPABASE_ANON_KEY	Enables Google sign-in
SESSION_SECRET	Required in production
APP_ENV	development or production
ANTHROPIC_API_KEY, ANTHROPIC_MODEL	Default model: claude-haiku-4-5
GOOGLE_SEARCH_API_KEY	Comma-separate multiple keys to rotate
GOOGLE_SEARCH_ENGINE_ID	Must search the entire web
GOOGLE_SEARCH_DAILY_CAP	Default 90

Supabase setup

• Tables: screenings, screening_files, job_events, job_runs
• Storage bucket: screening-files
• Auth: enable Google; add <your-url>/auth/callback as a redirect URL

Configuration

Scoring tiers, bands, partner lists and source toggles live in config.yaml.

Layout

app/
  main.py        routes and job runner
  auth.py        Google OAuth, signed sessions
  db.py          Supabase access
  pipeline/      scraping, search budget, LLM scoring
  render/        HTML templates, .docx and .xlsx reports
config.yaml      scoring rules

Deploy

Runs as a single long-lived process (jobs are held in memory), so use a container or web service, not serverless. render.yaml is included. OCR needs tesseract and poppler installed.

Limits

• One worker only.
• Public sources only; some companies simply publish nothing.
• AI-generated analysis. A human checks every figure before it leaves the building.

Research what can be known. Verify what cannot.