your verified score · everywhere you go
the verified proof-of-work index for tech talent.
Converts scattered evidence into a domain aware score, an evidence passport, and a path recruiters trust as a screening shortcut.
the candidate cannot prove they are good. the recruiter cannot tell who is.
Evidence is spread across a GitHub account, a portfolio domain, a Vercel deployment, a hackathon Devpost, three certificates in three issuer portals, a transcript, and a PDF resume that flattens all of it into bullet points anyone can fabricate. There is no single, trusted, portable artifact.
Screening means manually reconstructing the same four things every time — education, skills, experience, projects — from documents that are unverified by default. At volume this collapses into crude proxies: university brand, keyword matching, referral networks.
produce a verified, domain-calibrated evidence profile that a candidate owns and a recruiter can trust at a glance — and we sit between both sides of a transaction that is painful for both.
connect your evidence; get a score, a confidence level, an evidence passport, and a concrete plan to move the number.
careerscore — domain-adjusted, norm-referenced
verification confidence — weighted evidence at L2+
percentile — vs. new-grad ml profiles
figure 3.1 — two headline numbers (score + confidence), a detected domain, and a per-pillar breakdown where every bar is clickable down to its evidence.
a score people accept is not the one with the cleverest model — it is the one whose behaviour is principled and predictable.
The model only extracts facts and explains results. The score itself is deterministic TypeScript over numeric features. Same evidence in → same score out, every time.
We always emit Score and Verification Confidence separately. A 900 at 40% confidence and a 720 at 92% are different objects — confidence is what makes a score safe to share.
Prestige is capped; shipped, verifiable artifacts dominate. A candidate from a weaker school can — and should — outrank a prestige profile through real work.
There is no universal “tech” ruler. Each detected domain supplies its own weight vector and proof checklist, so the score measures the right things.
742 is meaningless in isolation — exactly like a raw SAT count. Every score is norm-referenced to a percentile within domain and career stage.
We launch a transparent heuristic, collect outcomes, then calibrate the weights against success. We never claim predictive power we have not earned.
the platform detects which tech you're in — then re-distributes the 1000 points to that domain's weight vector. toggle a domain and watch the ruler change.
research-tilted: education and momentum rise; published benchmarks and model cards count.
→ AI / ML — research-tilted
table 6.3 — illustrative domain weight matrix; every row sums to 1000. a backend engineer applying for ml roles can switch their target domain and see how they score on the ml ruler.
the product's entire moat is trust. a shared profile carries a cryptographic signature, so a recruiter knows it was not edited after the fact.
User typed “I interned at X.”
Internship certificate PDF on file.
GitHub repo / portfolio / hackathon page matches the claim.
GitHub connected via OAuth; account identity confirmed.
University email, company email, signed certificate, Credly badge.
An institution or employer partner confirms directly.
a self-claimed resume has zero marginal trust. a signed, evidence-graded, domain-calibrated profile has real trust — the only thing that lets an employer act on the output.
the product must feed hiring without becoming a regulated automated employment-decision tool. here is the staged path that captures the value and keeps the company defensible.
Private, candidate-facing score, passport and roadmap. Zero employment-law exposure. This is the entire MVP.
The candidate generates a signed verified profile and chooses to send it. Candidate-initiated, consentful — the recruiter gets pre-verified evidence.
Employers see ranked, domain-matched, evidence-backed candidates — but the tool never auto-rejects. A human decides. Full compliance scaffolding required before launch.
Talent pool, ATS & assessment integrations, partner API. Network effects compound. Only after Stage 3 is proven, audited and trusted.
Human-in-the-loop, never auto-rejects. Explainable, job-related, bias-audited — age and protected attributes are excluded as inputs.
facts are extracted by AI, verified, classified by domain, converted to numeric features, scored deterministically, normed to a percentile, and only then explained back in language.
CV, GitHub, links, certs → private storage
PDF / DOCX / image → text
Structured claims via constrained LLM
OAuth, issuer, cross-check → evidence level
Primary + secondary + confidence
Facts → numeric feature vector
Domain weight vector → pillars → total (deterministic)
Percentile vs. domain × stage cohort
LLM narrates score + roadmap
Passport, signed share, dashboard
deterministic / verifiable llm-assisted (extraction & explanation only)
the governing rule: self-claimed inflates nothing. the cheapest path to a higher score is to do verifiable work.
CareerScore turns scattered career proof into a verified, domain-aware score, an evidence passport, and an improvement roadmap — the trusted screening layer that saves recruiters the first hour on every candidate, and shows every student exactly where they stand.