Oddit is a dev evaluation engine that turns a candidate's real shipped code into a 5-minute hire memo with file-and-line evidence and 11 work-pattern signals on how they actually engineer.
Hiring developers in 2026 has broken in a specific way. Candidates use AI to write their resumes. Engineers use AI to write the code they ship. Greenhouse's 2026 AI in Hiring Report (n=4,136) found 86% of recruiters caught or suspected candidate fraud last year, and 42% of candidates admit prompt-injecting their resumes. Meanwhile JetBrains' January 2026 survey of 10,000+ developers shows 90% use AI tools at work to write code.
So hiring teams have never had more evidence about a candidate, and never had less signal about whether the work is actually theirs or how they actually think. Bersin pegs the all-in cost of one bad senior engineering hire at $375k to $500k. Hiring managers are spending an hour or more reading each candidate's code and projects before a final-round call, guessing at what's AI-generated and what isn't, and still getting it wrong.
We solve this by auditing the candidate's actual code and producing a hire memo in 5 minutes. The memo classifies what they wrote (custom vs framework glue vs AI-assisted), cites file-and-line evidence for every claim, and surfaces 11 work-pattern signals (rationale habit, iteration discipline, decision quality, scope management, and 7 more) so the hiring team sees not just what they shipped but how they actually engineer.
Built and shipped the engine. Working multi-model audit pipeline in production: Code Property Graph analysis across 15+ languages via tree-sitter, 17 deterministic pattern detectors, a tool-calling verification agent that re-checks every claim against the actual code, and 11 person-level work-pattern signals across all of a developer's audited work. 153 audits run on real developer code since the May launch. 77 users signed up.