AI Hiring Pipeline

The problem

Hiring pipelines leak at every handoff. A recruiter writes a post, sources candidates, screens resumes, schedules interviews, runs them, scores them, then decides — and each step is manual, subjective, and slow enough that good candidates go elsewhere.

We automated the whole chain.

The chain

1. AI-assisted job posting — structured role definition from a loose brief.
2. AI-assisted candidate outreach — sourcing and first contact.
3. Resume scoring — parse, normalise, score against the role.
4. Match ranking — candidates ordered by job-match score, not arrival time.
5. Personalised interview outreach — generated per candidate, per role.
6. Fully automated AI interviews — conducted without a human in the loop.
7. Scoring and shortlisting — with finalisation suggestions for the hiring team.

Each stage hands a typed artifact to the next, so a failure at step 4 doesn’t corrupt step 6 — it just stops.

The hard parts

One-way video interviews. I built the interview interface on WebRTC rather than renting one. Recording, resilience to flaky connections, and a UI a nervous candidate can use without instructions.

Proctored assessments. On-screen tests for enterprise hiring teams with the checks needed to prevent unfair means — without turning the experience into surveillance theatre.

Django backend, Bubble front-end. An unusual split. The backend carried all the logic and exposed a narrow API, which is what made the front-end choice survivable.