Insignia Recruit
Hiring intelligence, scored from public signal.
Insignia Recruit was the studio's first published product. A hiring engine that scored candidates against engineering competencies, pulling skill signal from public technical artifacts (commits, papers, talks) rather than CV keyword matching. The thesis was straightforward: skill-from-signal beats skill-from-claim. Recruit proved the thesis at small scale before studio focus consolidated to flagship products.
What it shipped
- 01Skill graph derived from public technical artifacts
- 02Calibrated scoring across roles and seniorities
- 03Bias-aware shortlisting with audit trails
- 04Native ATS integration with Greenhouse, Lever, and Workday
Why we built it
Senior-engineering hiring was being conducted on CV scans. Public signal, the kind that actually evidences competence (open-source commits, conference talks, technical writing), was abundant and largely ignored. We built Recruit to invert the priority. Public artifacts became the primary evidence; the CV became supporting material.
What we learned
- 01
Calibration is a discipline, not a feature.
We invested heavily in calibrating scoring across role bands and seniorities. The discipline compounded; once calibration was treated as a first-class production concern rather than a launch milestone, every downstream decision improved.
- 02
Audit trails on every shortlist decision.
Both for legal posture and for iteration. Recommendation-grade decisions that don't carry their reasoning forward become noise. We made every shortlist call traceable to the signals that produced it.
- 03
Bias-aware scoring is a permanent constraint.
Not a launch task to clear and forget. We treated bias-aware scoring as a permanent architectural constraint that gated every model change. The discipline carried into every recommendation-grade system the studio has shipped since.
What we'd do differently
- 01
Tighter integration with interview scheduling.
Recruit ended at shortlist generation; operators still handled scheduling and coordination manually. Closing that gap would have made the product feel like an end-to-end hiring engine rather than a discovery layer.
- 02
Monthly calibration, not quarterly.
Quarterly calibration was the convenient cadence; monthly would have caught role-band drift earlier. Calibration cadence matters more than calibration depth.
- 03
Less coupling to specific ATS vendors.
The Greenhouse / Lever / Workday adapters were maintenance-heavy. An export-first architecture (we generate artifacts; the ATS consumes them) would have aged better than three bespoke integrations.
Why it's retired
Studio focus consolidated to flagship products. Recruit's lessons were absorbed into our hiring practice and into the recommendation architecture we now ship in Auspex. The product was retired honorably with existing customer accounts migrated to alternative providers; data was returned to operators on request.
Where the lessons live now
- 01Calibration discipline
Every Auspex recommendation carries calibration metadata. The pattern originated here.
- 02Audit-trail-first design
The audit ledger architecture in Verum and Auspex traces its lineage to Recruit's shortlist audit trail.
- 03Bias-aware scoring as constraint
Now a permanent architectural posture across studio products, not a feature.