23 Apr 2026
Autonomous job agent (envoy) — full pipeline build
Built an autonomous job agent that handles the entire job search pipeline — so I don't have to.
It discovers listings across 20+ boards (SA-specific, global remote, freelance platforms), enriches company data, scores opportunities, auto-applies where possible, and monitors a dedicated Gmail inbox for responses. Full state machine from discovered → submitted → interview_scheduled → offer_received.
Some bits I'm particularly happy with:
Pre-hiring signals — monitors GitHub/GitLab org activity (new repos, team changes, releases) and drafts proactive cold outreach before a job posting even exists. If a company is clearly building, that's a better time to reach out than after 200 people have already applied.
Scoring isn't naive — weighs company health (stability, culture, funding stage) against job-level signals (salary, recency, remote). An LLM layer on top adjusts the final score based on tech stack fit, seniority match, and red flags buried in the JD.
Answer resolution has fallback tiers — profile first, then learned answers, then LLM match, then LLM generation. Anything it can't resolve goes to a human-in-the-loop queue rather than guessing.
Email monitor classifies inbound replies (interview, assessment, offer, rejection) and advances application state automatically.
Stack: Python 3.11, SQLite/WAL, FastAPI, Vue 3, Playwright for ATS automation, Ollama for local LLM inference, 2captcha for Turnstile.
Cyberpunk dark theme dashboard because obviously.
Still heavy WIP but the core pipeline is running. Happy to talk architecture if anyone's building something similar.