Case 01 / 04
AI / HR-tech
AI Interview Agent
Project brief
A voice AI that runs first-round candidate interviews, evaluates them with LLMs and hands employers a structured report.
The problem
First-round screening is repetitive, and two interviewers rarely judge candidates the same way. The platform runs the interview by voice, scores it with an LLM against the role’s criteria and gives the employer a report they can compare across candidates. Volume follows each employer’s hiring pipeline, so the system has to cope with quiet days and sudden batches alike.
What I delivered
- Interview orchestration with Vapi voice agents.
- LLM-based evaluation of every interview using OpenAI and Gemini.
- Structured employer reports in a Next.js + Supabase web app.
The reliability fix
Evaluations are long-running: transcripts are long and prompts are chained. They were triggered through webhooks with hard time limits, so longer interviews timed out and their evaluations never reached the employer.
I redesigned the workflows in n8n to run asynchronously: accept the request immediately, process in the background, and deliver the result when it’s ready, with each interview evaluated exactly once.
Result: 100% of evaluations now complete and reach the employer, whatever the interview length or batch size.
I wrote up the pattern in Webhooks time out. LLM workflows don’t care.
Stack
OpenAI · Gemini · n8n · Vapi · React · Next.js · Tailwind CSS · Vercel · Supabase