Description
Summary Overview I need an experienced engineer to build a self-hosted personal-operations AI agent on Hermes Agent (Nous Research, Python), using the Claude API as the model provider, with PostgreSQL as the datastore. It captures tasks from Telegram (and later SMS, Todoist, and email), uses Claude to turn messages into structured tasks, sends a scheduled daily digest, and syncs a shared Todoist project as a read/write surface. This handles personal admin including bills, so trustworthiness, credential isolation, and clean security practices are as important as shipping speed. You will build against test credentials and a test database only; production secrets are injected by me. Scope (fixed-price milestones) Core loop: Telegram capture → Claude extraction → Postgres task store → confirmation flow → scheduled daily digest → task completion/loop-close. Approval gating for anything money-related. Todoist two-way sync (mirror out + inbound webhooks) and email intake with attachment/bill parsing. SMS channel (Twilio) after A2P 10DLC registration clears. A detailed build spec will be shared after screening + a mutual NDA. Stack: Hermes Agent (or strong transferable experience with OpenClaw / Claude Agent SDK / similar), Python, Claude API, PostgreSQL, Telegram Bot API, webhooks, Todoist API, Twilio (later). Container/sandbox isolation experience required. Process: shortlisted candidates complete a small paid test task (~1–2 hrs) before any milestone is awarded.