Studio

SkillBridge

SkillBridge helps graduate students translate academic work into a reviewable, evidence-backed portfolio of workforce-relevant skills.

Explore itResearch prototype · Internal · Validation and job-explorer work ahead
SkillBridge evidence-review workspace showing a source document beside extracted skills, each backed by a quoted passage with accept and reject controls.
ScreenshotThe evidence-review workspace: a source document beside its extracted skills, each backed by a quoted passage and waiting for the user to validate or reject it. All names, filenames, and document text are synthetic demonstration content.

The problem

The language of academic experience does not map cleanly to the language employers use. Graduate students build sophisticated capabilities through research, teaching, and projects — then struggle to name them in workforce terms, and AI tools that do it for them usually cannot show their evidence.

The approach

Upload syllabi, CVs, research summaries, thesis abstracts, and portfolios; a locally hosted language model extracts candidate skills and preserves the exact source passages behind each one. Users accept, reject, edit, or remap every suggestion, organize validated skills into a clustered portfolio, and run configurable matches against the O*NET occupational taxonomy — with each match explained by which skills drove it.

Research basis

SkillBridge connects the lab’s graduate career-readiness research to a working system, and points at a measurement question the lab takes seriously: the extraction pipeline is an individual feedback tool today, not a validated research instrument. The path from one to the other — a human-coded gold set, calibrated no-match behavior, reproducibility controls, an instrumented syllabus pilot — is designed and deliberately not yet claimed.

What it does today

  • Uploads and parses academic and professional documents (PDF, DOCX, TXT) with asynchronous processing.
  • Extracts candidate skills with a locally hosted language model, preserving the source passage behind every suggestion.
  • Maps reviewed skills to the O*NET taxonomy by semantic search over 20,000+ embedded occupational concepts.
  • Supports accepting, rejecting, editing, and remapping every suggestion — the user validates, the model proposes.
  • Clusters validated skills into a hierarchical competency portfolio with cross-document evidence and development over time.
  • Runs configurable occupation match runs with explainable, skill-level results.

What it does not do

  • Not publicly deployed; development authentication only, no university SSO.
  • No job-posting ingestion, individual-job matching, gap analysis, or application-narrative generation — those are planned, not implemented.
  • No documented external pilot or formal usability study. Internal smoke testing reshaped the matching workflow, but that is not evaluated user research.

Inside the prototype

Document library showing uploads with ready and failed processing states.
ScreenshotThe document library with processing states — including failures shown as failures. Synthetic demonstration content throughout.
Human review queue for O*NET skill mappings with moderate-confidence matches flagged for checking.
ScreenshotThe human review queue for O*NET mappings: moderate-confidence matches are flagged as worth checking before accepting, not silently applied.
Skills inventory clustered into domains and competency areas.
Screenshot236 extracted skills clustered into a competency portfolio across six domains. Counts reflect the demonstration corpus, not a study result.
An occupation match result in dark mode showing matched skills, their O*NET importance, and the source document behind each.
ScreenshotAn occupation match run with explainable results: which skills matched, their O*NET importance, and the document each came from. Occupation matching exists; matching to individual job postings does not yet.

Evidence

Verified working prototype

241 frontend and 140 backend tests passing (13 skipped), an optimized production build, and a health check reporting PostgreSQL, Redis, MinIO, the local language model, and O*NET embeddings ready — verified when the studio package was prepared.

As of 2026-08-28

Real interface, sanitized data

Twenty screenshots captured from the running application by a scripted process that replaces user names, filenames, passages, and skill wording with synthetic demonstration content — reproducible from the repository.

As of 2026-08-28

Occupational data at scale

20,441 O*NET concepts loaded and embedded for local semantic matching in the development environment.

As of 2026-08-28

What we are not claiming

  • SkillBridge does not claim to accurately measure a person’s skills, to produce valid or unbiased measurements across populations, or to improve career outcomes. It surfaces evidence-backed candidates for the person’s own judgment.
  • It matches validated skills to occupations, not to individual job postings — the distinction between what exists and what is planned.
  • Local inference is a privacy posture, not a security certification; the prototype is not production-secure or institutionally deployable.

Responsible use, privacy, and rights

  • Uploaded documents, parsed text, source passages, and user profiles stay private; the demonstration corpus is synthetic.
  • Job-posting data (NLx) is not ingested or shown — storage and reuse rights require written confirmation first, and the planned corpus work needs an IRB determination before being described as a study.

More from the Studio

All entries

Use it

Course Sphynx

A faculty-facing platform for helping instructors redesign courses through guided, transformation-oriented support.

Explore it

GATOS qualitative analysis workflow

A research workflow for using open-source generative AI and machine learning to support inductive qualitative codebook development.

Explore it

CATS QDA

A local-first workspace where qualitative researchers analyze sensitive interview data with AI assistance they can inspect, override, and audit.