The team your brief deserves.
Ensemblr reads every CV in your firm, understands a project brief and assembles the team that fits. Every recommendation arrives with evidence.
Staffing was a thread. Now it is a brief.
- Jonas K.Mon 09:12Anyone free for a data platform thing from November? Nordic retailer, roughly four months.
- Louise D.Mon 09:40Katrine might be, but I think she's on the bank until December?
- Anna S.Mon 11:05Ask her team lead. Or the capacity sheet, if it's been updated.
- Engagement lead..
- Data platform engineer..
- Data scientist..
- Change consultant..
How Ensemblr
works.
// Process- 01
Ingest.
Drop in CVs as PDF, DOCX or PPTX. One profile schema comes out. Contact details never go in.
- 02
Brief.
Paste the project description as it is. Roles, seniority and dates are proposed where the brief is silent.
- 03
Match.
Retrieval narrows two thousand people to twenty per role. A model scores each one against a fixed rubric and cites the projects it relied on.
- 04
Decide.
See coverage, gaps and budget for the whole team. Swap a person, accept, and the decision is recorded with its evidence.
What's in a match.
Every match, the same standard.
- 01
Any CV format.
Two-column PDFs, Word templates and slide decks all end in the same fields.
- 02
Evidence, not vibes.
Every score points at real project rows. Quotes are checked against the source before they are shown.
- 03
Coverage and risk.
A requirement by member matrix shows who covers what, and where a team has a single point of failure.
- 04
Integrity check.
Hidden text and instructions aimed at the model are flagged at ingestion. Two independent judges must agree.
- 05
EU by design.
Data at rest in Stockholm. Row level security on every table. An append-only audit log.
- 06
Open by API.
Everything the screen can do, a system or an agent can do through REST or MCP, with org-scoped keys.
- 09:41:07ingestcv_14.pdf · contact details stripped
- 09:41:12run_matchbrief 2291 · 20 retrieved per role
- 09:41:17scoreKatrine P. · 87 · 3 projects cited
- 09:41:22integritytwo judges agree · delta 2
- 09:41:27accept_teamby H. Talent · 5 members
- 09:41:32auditrow appended · nobody can delete it
Compliance and governance by design.
Built with the EU AI Act and GDPR in mind from the first table. Data stays in the EU, contact details never reach a model, a person makes every final call, and every decision is logged.
Asked by CTOs, mostly.
What formats can it read?
PDF, DOCX and PPTX today, including two-column layouts that break naive text extraction. The model reads PDFs as documents, not as scraped text. New formats are an adapter, not a rebuild.
How does it decide who fits?
Two stages. Vector retrieval plus hard filters on availability, seniority and location bring the pool down to about twenty per role. Then a language model scores each candidate on skills, industry, seniority and availability, and must cite specific projects from the CV as evidence.
Can a CV trick it?
It can try. Ingestion flags hidden text and instruction-like passages. Scoring only accepts structured output, clamps scores, verifies every quote against the source, and runs two independent judges. Disagreement goes to a person.
Where does the data live?
Postgres, auth and file storage in Stockholm. Compute runs in Stockholm too. Model providers receive profiles with contact details already removed, under a data processing agreement with EU standard contractual clauses.
Who makes the final call?
A person, every time. The tool proposes a team and shows its work. Someone on the talent team accepts, swaps or rejects, and that decision is stored with the evidence it was based on.
Does it scale past one firm?
Yes. Every row belongs to an organisation and the database enforces the boundary. A new client is one row and their CVs. Their capacity tool and CV repository plug in as adapters.