Telegram-excerpt monitors group chats, detects actionable discussions via LLM, and auto-generates PRDs to your private chat. Bugs and feature requests no longer get lost in noise. One bot per group → Firestore buffer → 3min silence triggers LLM classification → actionable threads become PRDs → delivered asmd files. GCP free tier, any OpenAI-compatible API, 22 classification scenarios. Open source, MIT licensed, self-hosted—your data stays private.

AppSignal — Get full visibility into app health, errors, and performance

Get full visibility into app health, errors, and performance

Hey everyone! I built this because I manage multiple Telegram groups for engineering teams, and I was drowning in messages trying to figure out what actually needed action. The key insight: most group conversations are noise. But buried in there are real bug reports, feature proposals, and technical decisions that should become tickets. I needed something that could tell the difference automatically. The classification prompt was the hardest part — it took 22 hand-crafted scenarios to get the LLM to reliably distinguish "the login endpoint is broken" from "lol who broke prod again" (with emoji, in Italian). The evaluation harness is included so you can tune it for your own team's communication style. Would love feedback on the approach. What would you add?