AI Job Autoresponder for 13 Sources with Telegram Approval

One freelancer cannot read a dozen marketplaces twice a day, spot duplicates, price each job and write a tailored proposal within platform limits. I built a single Python service with a clear pipeline: pre-filter, cross-platform duplicate check, Claude assessment, decision policy, pricing, then a queue or a Telegram card. Public feeds (forum RSS, Reddit, Hacker News, a public jobs API, alert e-mails over IMAP) are read through an isolated proxy reader; logged-in marketplaces through the user’s own browser session or an official MCP server. A Telegram bot with a team panel shows each card with the price, the proposal text and translations, and Send is the only approval. Per-platform pacing, daily caps, cooldowns after rate limits and a strict no-resend-after-click rule are built in. 13 job sources, 1,258 offline tests, 32 CLI commands, 12 architecture decision records.