Why this domain
Career transitions are exactly the kind of decision where a synthesis layer earns its keep. The noise-to-signal ratio in career media is high — listicles, anecdotal advice, sponsored content — and there's no liquid prediction market to defer to. The structured signals that exist (O*NET skill vectors, BLS wages, posting volume) are public but scattered, and rarely combined in a way that's legible to an individual making a real choice.
The research question scales cleanly from the business surface: can structured synthesis materially improve a job seeker's decision quality, relative to unaided search? Decision quality here is calibration of the seeker's confidence, regret after the move, and time-to-clarity. Forecast accuracy is not the metric. Match quality is.
How it will work
- 01Intake
Current role (O*NET SOC code or free text), education level, skills, ZIP, and your goal — earnings, similarity, or availability. Resume upload optional.
- 02Score
A pure scorer ranks candidate occupations by a weighted blend of earnings (BLS OES median wage), skill-vector similarity (O*NET), and demand (CareerOneStop posting volume). Outputs are z-scored against a sender-scoped candidate set.
- 03Advise
An LLM reads the top-ranked candidates, the intake, and the scorer rationale, and writes a structured recommendation: which transition is most defensible and why, drawing only from the data the scorer saw.
- 04Decide
You see the ranked list, the rationale, and the underlying signal — wage, similarity, demand — broken out per option. The product is the structured read; the decision is yours.
Sources
Public reference data, all already ingested into the system. None of it is proprietary; the value is in the synthesis.
- O*NET — Occupation taxonomy, skill importance vectors, typical education, STEM flag.
- BLS OES — Median annual wage and employment count per SOC code, refreshed annually.
- BLS Employment Projections — 10-year projected growth per SOC, used as an optional weight.
- CareerOneStop — Posting count and sample postings cached nightly per SOC.
What's shipped, what isn't
Shipped
- Reference data tables and ingestion jobs (O*NET, OES, BLS, CareerOneStop).
- Scorer (
server/careers/score.js) — pure, deterministic, tested. - Recommender orchestrator and LLM advisor (
server/careers/recommend.js,advisor/advise.js). - Intake / recommendations / snapshot schema with RLS policies.
Not yet
- Public API endpoints for intake submission and recommendation retrieval.
- Intake form, candidate-list rendering, advisor rationale UI.
- Pilot cohort onboarding (Albert Qian's network).
- Snapshot share URLs (parallel to the business surface's score snapshots).
Pilot cohort
Initial validation is planned with a small pilot group rather than a public launch. If you're a job seeker actively considering a transition and want to be on the pilot list, write to [email protected] — happy to share the framing in more detail.