Case Studies
Real work, including the parts that went wrong.
Three projects with the numbers attached. Each one ends with what it cannot claim — because a case study that only lists wins is marketing, not evidence.
A matching engine over 400,603 job postings
A national job board reduced to one shared skill vocabulary, so a résumé can be scored against 400,000+ postings with a reason code behind every point.
400,603
postings indexed
1.44M
extracted skill records
0
third-party dependencies
A fine-tuning pipeline, and the audit that caught it
A modular LoRA training framework — and the adversarial audit of our own work that found 84 issues, including the reason the model recited instead of reasoned.
84
issues found in our own pipeline
34/35
high-severity findings confirmed on re-check
1
finding refuted as a false positive
Two native apps over one deterministic core
A native iOS and Android platform where the shared code is strictly the deterministic engine, every pixel is platform-native, and no rule is written twice.
2
native front-ends
1
deterministic engine
0
reimplemented rules
Client engagements are covered by NDA and are not published here. Everything above is our own work — which is why the numbers and the failures can both be shown.
Related
semantic rankingContinue with another relevant resource.
- Related resourcecosine match 0.67Selected work: built, tested, measuredClosest content match: Selected work: built, tested, measured.Open resource
- Case studycosine match 0.59A fine-tuning pipeline, and the audit that caught itClosest content match: A fine-tuning pipeline, and the audit that caught it.View case study
- Case studycosine match 0.54A matching engine over 400,603 job postingsClosest content match: A matching engine over 400,603 job postings.View case study