← Back to Blog
Side Project

Building Hermes — an AI-Powered
Job Search Automation System

A personal project to automate the boring parts of job hunting using multiple AI agents, a shared memory layer, and smart model routing — so the human only shows up when judgment is needed.

March 2026
🤖

Hermes / Largeman

An always-on Telegram bot that runs the discovery pipeline, manages job state, generates resumes, and answers questions via chat.

💡

Claude / Cowork

Desktop planning sessions for architecture decisions, spec writing, and implementation planning. Works alongside Hermes, not in isolation.

🧠

Honcho Memory

A shared memory layer that keeps both agents in sync. Hermes reads context at session start; Claude writes summaries at session end.

OpenRouter

A single AI gateway that routes tasks to the right model — Claude Sonnet for writing quality, GPT-4o Mini for speed and cost, Haiku for quick checks.

Job Search Pipeline

Every job goes through the same structured flow — from automated discovery to final application. The human steps in only at the Approve stage, spending a few minutes per day reviewing what the system surfaced overnight.

Hermes Job Search PipelineDISCOVERYPREPARATIONAPPLICATION🔍Discover65+ ATS queries🧠AnalyzeLLM scoringApproveHuman review📄Tailor ResumeClaude Sonnet📊JobscanATS match score✉️Cover LetterAI-generated🚀AppliedDrive + Sheet loggedHuman-in-the-loop at Approve stage · All AI calls routed via OpenRouter
7Pipeline Stages
1Human touchpoint
~$0.05AI cost per job
AutomatedEverything else