Skip to content

CLAUDE.md

Latest commit

 

History

History
51 lines (40 loc) · 1.99 KB

CLAUDE.md

File metadata and controls

51 lines (40 loc) · 1.99 KB

HateWatch - Project Memory

Project Overview

HateWatch is a hate speech intelligence platform that monitors Telegram channels, scores messages for toxicity using Google's Perspective API, detects spikes in hate speech, and provides a dashboard for journalists and fact-checkers.

Tech Stack

  • Backend: Python 3.13, FastAPI, SQLAlchemy, SQLite
  • Frontend: React, Recharts
  • APIs: Telegram (Telethon), Google Perspective API
  • Scraping: Telethon for Telegram channels
  • Hosting: Render (backend), Vercel (frontend)

Key Files

  • scraper/telegram_scraper.py - Fetches messages from Telegram channels
  • processing/perspective.py - Scores toxicity via Perspective API
  • analysis/spike_detector.py - Detects unusual toxicity increases
  • api/main.py - FastAPI backend
  • dashboard/src/App.jsx - React dashboard

Live URLs

Commands

# Start API server
uvicorn api.main:app --reload --port 8000

# Start dashboard
cd dashboard && npm start

# Run scraper
python scripts/run_scraper.py --continuous

# Run processor
python scripts/run_processor.py --continuous

Recent Sessions

Session: 2026-01-30

  • Machine: MacBook
  • Pattern: DevOps / Deployment
  • Files changed: 12
  • Notes: Deployed HateWatch to production. Backend on Render (hatewatch-api.onrender.com), frontend on Vercel (dashboard-bay-tau.vercel.app). Added Dockerfile, nixpacks.toml, render.yaml, Procfile for deployment configs. Created /api/seed-demo endpoint for seeding demo data. Added "Last updated" timestamp to dashboard footer.

Session: 2025-12-24

  • Machine: MacBook
  • Pattern: Full Stack Development
  • Files changed: 46
  • Notes: Built complete HateWatch MVP from spec - Telegram scraper, Perspective API integration, spike detection, FastAPI backend, React dashboard with country filtering. Added Claude Code skills for /scrape, /process, /detect, /status, /report commands.