The goal was to build a narrow, dependable workflow around one specific operational problem:

• read job-related emails from my inbox
• identify which ones actually matter
• classify them into meaningful application states
• persist that state safely
• update the matching job record when the email clearly supports it
• give me a concise run summary I can check from my phone

Just as importantly, I wanted the system to be careful about what it did not do.

This project was not meant to be:

• a general-purpose inbox assistant
• a universal CRM sync engine
• a fully autonomous recruiting workflow
• a broad AI agent that takes uncontrolled actions

The target was much narrower than that.

I wanted something reliable enough to trust with real inbox traffic, but still small enough to reason about, test, evaluate, and improve without turning it into a large product.