Skills extracted from what a candidate built. Not from what they wrote.
The AI reads the actual files — code, thesis, design, reports — and every skill it extracts names the project it was read from.
Not Another Job Platform.
Artifact in, structured skills out
Four stages, fully automated. The candidate uploads, the AI extracts, the platform shows evidence, the recruiter searches. No manual tagging.
Upload
The candidate uploads project files: source code, thesis PDF, design files, internship reports, lab notebooks. One paragraph of description in the candidate's own language.
Extract
The AI reads the artifacts. It parses code, summarizes documents, transcribes video walkthroughs. It produces a structured skill list across five typed buckets.
Evidence
Each extracted skill is stored on the project it was read from, together with the evidence sentence the model wrote for it, where it gave one. The recruiter sees the skills grouped by project, next to the list of files that were analysed.
Search
Recruiters search by skill, role, location, or in natural language. Matches are ranked on the extracted evidence, not on keyword overlap with the CV.
Five typed skill buckets
Skills are never a flat list. Every extraction categorizes into the same five buckets, so a recruiter searching for soft skills doesn't have to wade through SQL queries.
Hard skills
Programming languages, frameworks, databases, tools. Extracted from code repositories and technical artifacts.
Soft skills
Communication, collaboration, project management. Extracted from documentation quality, README structure, internship evaluations.
Design skills
UI/UX, mechanical CAD, architectural drawing, graphic design. Extracted from design files, sketches, prototypes.
Domain knowledge
Industry-specific competence: finance, biotech, manufacturing, education, public sector. Extracted from project context and applied work.
Languages
Working languages the artifacts themselves are written or delivered in — a thesis, a report, a README, a deck. Extracted from the file, never from a declared or certified level.
One README in, five typed buckets out
An illustrative example, not a live extraction: a sample project README on the left, the five typed buckets on the right.
# Adaptive reuse of the ex-Falck industrial site Master thesis · Politecnico di Milano · Architecture · 2025 Designed a mixed-use intervention on 12,000 m² of the former Falck steelworks in Sesto San Giovanni. The plan keeps three of the original blast-furnace structures (listed under Italian heritage law D.Lgs. 42/2004) and inserts new residential, retail, and public-space programs around them. Deliverables: 1:200 master plan, 1:50 sectional studies of the heritage shells, full sustainability assessment under LEED v4 BD+C, structural feasibility study with FEM analysis on the reused frames. Software: Revit (BIM model), AutoCAD, Grasshopper + Rhino, Adobe InDesign for the boards. Worked with two structural engineers on the FEM model and with the municipality's heritage office on the listed-element constraints. Final review: presented to the jury in English; final PDF set in Italian for the regional planning authority.
In the product, where the analysis cited evidence, the sentence it found is shown with the skill.
Every extracted skill names the work it came from
We don't ask anyone to trust the AI on its word. A recruiter opens a project on the candidate's profile and sees the skills listed for it, next to the names of the files that were analyzed.
- A verification tier on every skill — Verified, AI-extracted or Self-reported — shown to recruiters
- The project each extracted skill was read from, with the analysed files named when the student shares them
- Public algorithm registry with a card for the talent-match model that ranks candidates for a role
What the AI can't do
We're transparent about the limits because trust comes from honesty about both. Here's what artifact-based extraction is not.
- It doesn't measure a candidate's potential — only the skills evidenced in what they've already built
- It doesn't replace the interview — it removes the screening step that should never have been an interview
- It doesn't certify the work was the candidate's own — that's where the optional professor endorsement adds a trust layer
Search candidates on evidence, not on claims.
Free to browse the verified pool, contact 5 candidates per month with a corporate email. Try the AI search before signing up.