Castr · The engine behind Zero Gravity Staffing
How we removed
Gravity from Staffing.
We rebuilt our own staffing operation around one engine: Castr ranks every candidate, tailors every CV, and tells you who's actually available — in seconds, all explained. Now we're handing it to other agencies, ready to activate.
30-day pilot · self-hosted optional · live in 48 hours
Match · #047
02:14 ago
Shortlisted candidate → Senior Data Scientist
92
score
- Skills match94
- Domain experience88
- Cultural signals86
- Knockouts cleared100
Why this match
Strong on PyTorch + LLM eval (3 yrs at scale). Lighter on MLOps but mentored two juniors — counts as a partial signal.
650+
candidates parsed daily
in production
92%
top-3 match accuracy
vs human reviewer
live
availability state
per candidate, per day
3.4×
faster shortlisting
median across teams
Inside the app
Open the job.
The shortlist is already there.
Paste a job, drop your CV corpus, and Castr ranks the entire pool in seconds — with a reason on every row. No spreadsheets. No 14-tab kanban.
app.castr / matches / senior-data-scientist
247 ranked · Berlin · €85–110k
- 01
Candidate · #C-2184
8 yrs ML · scale-up SaaS
94 - 02
Candidate · #C-1907
6 yrs data science · EU bank
91 - 03
Candidate · #C-2261
Lead DS · industrial group
88 - 04
Candidate · #C-1843
ML lead · contract · public sector
84
Every rank carries a reason · click a row, see the trail
How matching works
Five passes.
One ranked list.
Most ATSes do keyword search and call it AI. Castr runs every match through a five-stage pipeline — each stage does what it's good at, no more.
Ingest
Extracts skills, dates, employers, education and languages from PDFs, DOCX, LinkedIn dumps — into a structured profile. Same for the job: paste a JD or URL, get a parsed spec.
Knockout
Non-negotiables first: clearance, location, languages, work permit. Cheap, explainable, and they protect the rest of the pipeline from noise.
Pre-rank
Structured signals — years of experience, role family, employer tier — produce a fast first ranking. Most of the corpus is filtered out here, cheaply.
Vector
Embeddings catch what keywords miss: adjacent skills, transferable experience, the candidate your filter rejected who turns out to be the hire.
LLM eval
Only the shortlist reaches the LLM. Fit/gap reasoning in plain language — why they ranked here, what's a stretch. Audit-logged, weight-aware, defensible.
Castr answers with context, not lists. Decisions, outcomes and the metrics about them live in one stream.
Beyond matching
Matching is just the start.
Castr keeps your talent pool ready.
Castr parses every CV into structured data, re-shapes it to any opportunity in any format, and tracks who's actually available — so nobody is reformatting Word docs at 11pm again.
CV auto-adaptation
opportunity #2041
One CV in.
Every format out.
Castr parses each CV into a structured candidate record — skills, roles, dates, outcomes — stored in your database, not in a Word file. From there it rewrites and re-renders the CV for any opportunity, in any client template. No copy-paste, no manual reformatting, no version drift.
One source of truth · re-rendered on demand
Availability management
week of 20 May
Who's actually
ready today.
Bench, rolling off, passive, on assignment, off-limits — Castr tracks the state per candidate and updates it from your calendar, project pipeline and conversations. The shortlist already knows who you can actually call.
Triggers · Slack · email · pipeline event · calendar change
New opportunity? Castr tailors the CVs of the top matches and pings the recruiter with who's free. No re-typing. No "who's in the pool?" thread.
A workspace built for recruiters
Built for the way
recruiters actually work.
01
Ranked matches
Every job has a live, weighted shortlist. Strong fit, lateral move, stretch — each candidate carries a score and a reason.
02
AI Chat
Ask in plain language: "Find me Python engineers in Madrid who've shipped LLM features." Castr knows your data.
03
Explainable scores
Every rank comes with a fit/gap reasoning trail. No more "why is this person at #3?" — the answer is in the row.
04
Audit-ready
Every prompt, model, score and decision is logged. GDPR and EU AI Act ready, with a pre-filled Annex IV template at signup.
05
Self-measuring
Every match, override, placement and rejection is captured by the same engine that ranked them. Accuracy, time-to-shortlist and weight drift are read off the workflow — not rebuilt in a separate analytics pane.
06
Bring your own weights
Set weighted scoring profiles per role family — skills, experience, education, languages, knockouts. The LLM tier respects them, and the next run uses what the last one learned.
07
Your data, your tenant
Sold standalone per agency. AI calls route to Mistral, OpenAI or Gemini (configurable). EU customers default to EU-hosted Mistral.
Recruitment AI is high-risk under the EU AI Act · Castr ships audit logs, human-in-the-loop, candidate-visible explanations and a documented pipeline — compliance-ready day one.
Voices from the field
Recruitment teams
who care about precision.
Anonymised feedback from agencies running Castr in pilot and production. Roles and sectors only — no client names.
We were spending six hours per role on shortlists. Castr does the first pass in eight minutes — and the explanations mean my team actually trusts it.
Head of Talent · mid-size Benelux recruitment agency
Castr caught a candidate our keyword filter rejected — turned out to be the eventual hire.
Founder · boutique executive-search firm
The audit log alone is worth it. Every decision is defensible to clients and to the AI Act.
Operations Lead · IT staffing agency
VI · Launch
Stand up Castr on your
own data. In 48 hours.
Bring your CV corpus and a live job. We provision your tenant, import your data, and you watch Castr rank, tailor and surface availability on your own talent pool — not a generic demo deck.
Self-hosted · €490/mo · migration included · cancel anytime in pilot
Your first 48 hours
- 01Day 0 · tenant provisioned, SSO wired
- 02Day 1 · your CVs parsed into structured records
- 03Day 1 · live jobs ranked with reasoning
- 04Day 2 · CV templates + availability triggers live
- 05Day 2 · audit log + AI Act posture handed over