chitti-news

name: chitti-news description: Chitti News — state-aware multi-language Indian news aggregator. Aggregates 25+ RSS feeds across English and Hindi (regional languages stubbed for v1.1), serves articles by state × language × category, renders DeepSeek-powered "Chitti's Take" 3-bullet summaries, runs a fact-checker that cross-references ≥2 sources, and offers Read Later / Cancelled folders per device.

Chitti News — top-level skill

This skill describes the Chitti News product as a whole. It loads when the user asks about Chitti News, news aggregation, RSS feeds, or any of the per-category sub-agents below.

Repo layout

chitti-news/
├── frontend/index.html         mirror of workspace-root chitti_news.html
└── backend/
    ├── main.py                 Flask app · CORS · startup hooks
    ├── config.py               settings (DATABASE_URL · DEEPSEEK_API_KEY · CORS · scheduler)
    ├── database.py             SQLAlchemy engine + ensure_schema('news')
    ├── requirements.txt        flask · feedparser · sqlalchemy · httpx (DeepSeek REST) · rapidfuzz · apscheduler
    ├── runtime.txt + .python-version  → Python 3.11
    ├── render.yaml             Railway Blueprint
    ├── data/
    │   ├── sources.json        25+ RSS feeds across state × language × category
    │   └── articles_seed.json  6-row welcome seed (EN + HI)
    ├── models/                 Article, Source, ReadLater, BreakingAlert, FactCheck
    ├── services/
    │   ├── news_seed.py        loads JSON seeds on first boot (idempotent)
    │   ├── news_db.py          read paths: feed(), list_breaking(), get_article()
    │   ├── news_ingest.py      feedparser-based RSS poller (per-source, idempotent on link)
    │   ├── news_summary.py     Chitti's Take — DeepSeek 3-bullet, falls back to RSS summary
    │   ├── news_factcheck.py   cross-source fact-check (rapidfuzz title-similarity, cached 6h)
    │   └── news_scheduler.py   APScheduler: rss_poll every 30 min · daily_breaking 06:00 IST
    └── routes/
        └── news.py             Blueprint /api/news/*

Endpoint surface (Flask Blueprint, prefix `/api/news`)

Method	Path	Purpose
GET	`/feed?state=&language=&category=&limit=`	Main feed
GET	`/<state>/<language>/<category>`	Pretty alias (e.g. `/india/hi/business`)
GET	`/breaking`	Active breaking-news alerts (≥3 sources agree)
GET	`/article/<id>`	Single article
GET	`/article/<id>/take?language=`	Chitti's Take (3 bullets via DeepSeek)
GET/POST	`/article/<id>/factcheck`	Cross-source verdict (cached 6h)
GET	`/sources?state=&language=`	Source registry slice
POST	`/save`	Add to `saved` or `cancelled` folder (X-User-Token header)
GET	`/save?folder=`	List a folder
DELETE	`/save/<entry_id>`	Remove from folder
GET	`/scheduler/status`	Diagnostic
POST	`/scheduler/trigger/<job_id>`	Force-run a scheduler job

Schema isolation

All News tables live under the news schema in the shared Supabase Postgres (alongside medupi.* and shares.*). See backend/models/_schema.py. Schema is auto-created on startup via database.ensure_schema().

Sub-agents

The .claude/skills/chitti-news-*/ directories define per-category agents with their own context + tone:

chitti-news-summarizer — Chitti's Take (3-bullet AI summaries)
chitti-news-factcheck — cross-source verification
chitti-news-politics — political news context + neutrality guardrails
chitti-news-sports — sports news + scores + schedules context
chitti-news-business — business news + market context
chitti-news-tech — technology news + product launches context
chitti-news-entertainment — entertainment news + reviews context

When the user asks a category-specific question, prefer the matching sub-agent over the generic chitti-news skill.

Live URLs

Frontend: https://sahayai.in/chitti_news.html
Backend: https://chitti-news-api-production.up.railway.app (planned — render.yaml ready)