JobSignal — HN Job Board - 图片 1

JobSignal — HN Job Board

HN Who’s Hiring → structured jobs with schema/verifier & eval gates · jobsignal.dev

Structure PipelineSchema + VerifiersGolden-set EvalPythonNext.jsTurso

1.The Challenge

Hacker News Who’s Hiring posts are unstructured and noisy. Manual scanning is slow, and naive LLM extraction hallucinates fields while burning API cost — seekers need trustworthy structured filters, not another raw thread dump.

2.The Solution

Built a code-owned production path (fetch → structure) without LangChain: rules and pre-filters, StructureAgent on Groq with model fallback, schema validate/repair, then verifiers before write. Golden-set eval gates on every PR track field accuracy (~93% → 97% → 97.3% → 100% on a 16-sample set). User job_corrections feed a verdict step and few-shot exemplars back into structure. Next.js product on Turso (SQLite) with filters, profile/fit, and company pages. Earlier research/match/decision agents remain scaffolded, not in the daily path.

3.The Impact

Live at jobsignal.dev: structured browse/filter, explainable Best Fit from local profile signals, and company context that raw HN threads cannot provide. Corrections + daily verdict + few-shot close a measurable quality loop alongside golden-set PR gates.