🤖 AI Intelligence Daily

A standalone Telegram bot that delivers a deep, founder-focused AI intelligence brief every morning at 06:30 IST (01:00 UTC).

Every day it fetches the last 24 hours of AI news, research, and tool releases from ~25 RSS sources, processes everything through Groq LLaMA 3.3-70B, and sends a structured 5-section brief to Telegram — followed by 2 researched build suggestions with 👍/👎 buttons, in the same message.

Sister project to Daily-Digest NewsBot. This is a separate, standalone bot with its own Telegram bot token. It reuses the same fetching / dedup / Telegram patterns but is focused entirely on AI intelligence for a solo founder building D2C brands and AI tools in India. The two projects do not share code or runtime — only conventions.

The 5 sections

1. 🚀 NEW RELEASES — model launches & product drops (WHAT / WHY / FOUNDER USE CASE / DIFFICULTY)
2. 🧠 RESEARCH WORTH KNOWING — practical research only (FINDING / WHY / WATCH FOR)
3. 🛠 TOOLS YOU CAN USE TODAY — immediately usable tools (TOOL / BEST FOR / TRY IT)
4. 💰 AI BUSINESS — funding, acquisitions, deals (WHAT / SIGNAL / INDIA ANGLE)
5. 💡 FOUNDER INSIGHT — one sharp, actionable takeaway

🧭 BUILD NEXT — the suggestions section

Appended to the same daily message (never a second one). Exactly two suggestions, in two fixed slots, each written as What / Why now / First step.

Slot 1 — #prism-glamshelf, every day. Pulls open issues from Prism-Productivity-Tool and glamshelf-twin, ranks them most-commented first with the oldest breaking ties, and builds the suggestion on the winner. Issues already used as a source recently are skipped so the slot does not stall on one ticket. When neither repo has a usable open issue, it falls back to a context-grounded idea for whichever project is more due — alternating between the two — so this slot is never empty.

Slot 2 — alternates #portfolio ⇄ #new-build. The flip is read from the database (the most recent slot-2 row), not from the date, so a skipped or failed run cannot desync the rotation.

• #portfolio — projects that strengthen an AI-engineering portfolio.
• #new-build — standalone tools, biased toward evidence of unmet demand (Reddit "somebody make this" / "I wish there was" posts) over trend-following. Hacker News is pulled in only as a lighter secondary signal.

Dedup. The last ~20 titles go into the prompt as an exclusion list, and a SequenceMatcher guard rejects any generated title ≥0.75 similar to a recent one and regenerates once. The prompt asks; the guard enforces.

Feedback loop. Each suggestion ships with 👍/👎 inline buttons whose callback_data carries its row uuid. Taps are recorded by a separate Vercel function (ai-daily-reactions-webhook) — deliberately its own repo and deploy. The next morning's run reads the last ~10 reactions and feeds them into the generation prompt as calibration.

If Gemini, Supabase, or GitHub is down, the suggestions section is skipped and the news brief still ships. Setting SUGGESTIONS_ENABLED=0 does the same thing deliberately.

File structure

AI-Intelligence-Daily/
├── bot.py              # Entrypoint, --run-now flag; assembles the single message
├── fetcher.py          # RSS fetching (requests + browser UA + feedparser), 24h filter, dedup
├── summarizer.py       # Groq API call with the deep founder prompt
├── telegram_sender.py  # Telegram send with 4000-char chunking + inline keyboard
├── config.py           # RSS feeds, system prompts, env vars
├── dedup.py            # 24h headline dedup tracker
├── suggester.py        # BUILD NEXT: slot logic, Gemini call, title dedup guard
├── github_issues.py    # Slot 1 evidence — open issues from the two repos
├── demand_signals.py   # Slot 2 evidence — unmet-demand posts for #new-build
├── supabase_client.py  # PostgREST reads/writes for suggestions + reactions
├── sql/001_suggestions.sql  # Schema for the two tables (run once)
├── sent_stories.json   # Dedup state (committed by the workflow)
├── requirements.txt
├── .env.example
├── .gitignore
├── Procfile
└── .github/workflows/ai_daily.yml

Local setup

pip install -r requirements.txt
cp .env.example .env        # then fill in your keys
python bot.py --run-now     # build + send one brief immediately

Then apply sql/001_suggestions.sql once against your Supabase project.

Environment variables

Var	Notes
GROQ_API_KEY	Same key as the NewsBot is fine
TELEGRAM_BOT_TOKEN	Must be a NEW bot (e.g. @AI_Intel_Daily_bot) — do not reuse the NewsBot's token
TELEGRAM_CHAT_ID	Same chat ID as the NewsBot is fine
GEMINI_API_KEY	Free tier. Generates the two daily suggestions
SUPABASE_URL	Project that holds suggestions + reactions
SUPABASE_SERVICE_KEY	Service role — both tables are RLS-locked with no policies
GITHUB_TOKEN	Reads open issues. Optional for public repos, but avoids the 60 req/hr anonymous limit
SUGGESTIONS_ENABLED	Optional. 0 ships the news brief without the BUILD NEXT section
GEMINI_MODELS	Optional. Comma-separated fallback order; first model that answers wins

Deployment

Triggered natively via GitHub Actions schedule at 06:00 IST (00:30 UTC daily) and via repository_dispatch from cron-job.org. See the workflow at .github/workflows/ai_daily.yml.

The workflow restores sent_stories.json from main, runs the bot, and commits the updated dedup state back so stories aren't repeated day to day. Suggestion history lives in Supabase, not in the repo.