Five layers. Three human gates. Discovers opportunities, researches companies, generates tailored documents — and never auto-submits.
Mass-apply tools send hundreds of generic applications. This system sends one — researched, tailored, and reviewed by a human before submission.
Click each stage to see what happens to a real listing as it moves from discovery to delivery.
Each provider contributes a different dimension. Click a provider to see its contribution to the Harnham briefing.
Real artifacts from the pipeline. Redacted, but structurally intact.
Structured, queryable knowledge base. Ten JSON files — skills, experience, achievements, education, narrative — each with strict Zod schemas.
Multi-turn AI agent loop producing company intelligence briefings. 30–40 reasoning steps per run across 8+ MCP sources.
Three-phase pipeline: content selection, parallel generation, multi-perspective audit. Zero tools, single-turn, deterministic.
Daily cron pipeline scanning four boards. Normalises, deduplicates, pre-filters, then scores via LLM.
Event-driven Telegram bot tying layers together. Background workers, PID tracking, Cloudflare KV, React dashboard.