Skip to content
focoit.com
Back to the lab

AI engineering · Applied AI

Jobflow

A job-application pipeline that scores fit with a language model and refuses to cite evidence it can't ground.

Diagram of Jobflow's two safety layers: names redacted before the hosted model, and evidence below a grounding threshold refused.

Capture job postings, keep a candidate profile, and let a language model score fit and match the skills a posting asks for. Profile evidence comes back through semantic search, with a citation for every claim.

Two safety layers wrap every AI call. Candidate and employer names are swapped out before anything leaves for a hosted model, and restored afterwards. Retrieval refuses evidence below a measured similarity threshold, and generated claims are checked against their citations.

It is multi-user, with invitation-only sign-up and roles. In production it runs on Azure OpenAI through a managed identity, with no API key.