Telegram Feedback Bot with AI

About This Project

A privacy-preserving feedback pipeline: bot intake in Telegram, categorized searchable records, moderation, and daily AI sentiment summaries cited back to source feedback.

Feedback arrived through community channels, social, support tickets, and scattered chats. Detecting recurring pain points, sentiment shifts, or product opportunities required manual triage.

Sentiment trends from the pipeline feed community health tracking and backlog priorities. Daily and weekly summaries list recurring issues without manual report writing.

Telegram bot intake → web UI → AI sentiment summaries that drive backlog priorities.

What I Worked On

System

Structured categories — ideas, fixes, developer questions (AMA mode) — plus temporary campaign-specific inputs; the bot speaks EN, RU, UK, and PL.

Privacy by design: players get a generated pseudonym (adjective_animal, ~10,700 combinations) instead of a visible Telegram ID; the ID-to-pseudonym mapping lives in a separate admin-only table, and anonymity is the player's choice.

Anti-abuse guardrails: daily submission limits, a 30-minute edit window, length validation, and text-only intake — all Redis-backed with graceful fallbacks.

Web app: dashboard with category/date/pseudonym filtering, text search and pagination, AMA question view, summary history, blacklist management, and a REST API with health checks.

Scheduled AI summaries (three per day plus weekly) through GPT-4o-mini behind a LiteLLM proxy, with stats and citations back to the source feedback.

Moderation from chat or web: ban/unban by pseudonym or ID with stored reasons; Slack Block Kit notifications deep-link every submission; optional Google Sheets mirror.