Industry
AI
Case Study
AI Job Hunting Platform
A résumé says where someone worked. It doesn't say how they talk under pressure. At brigit.dev I worked on voice interviews, fit scoring, and check-ins after someone actually starts, when most companies go quiet.
Project Overview
Hiring teams look at CVs, run a few interviews, make an offer, and then find out three months later the person doesn't fit. I worked on a product at brigit.dev that tries to catch that earlier, through AI conversations, team-fit scoring, and check-ins after someone actually starts.
I owned the whole thing. The voice interviews, the hiring pipeline, the scoring, the backend tying Python AI services to the web app.
Hiring product that talks to candidates, checks team fit, and follows up after someone joins, not just reads résumés.
Services
- UX/UI Design
- Web Development
- Product Strategy
Key Deliverables
- Responsive Design
- Custom Development
- User Experience Improvements
Platform Type
Web Platform
Tech Stack
- TypeScript
- Next.js
- Hono
- Python
- Pipecat
- Daily
- tRPC
- PostgreSQL
- Drizzle ORM
The Challenge
A résumé tells you where someone worked. It doesn't tell you how they talk, how they handle pressure, or whether they'll mesh with the team they're joining.
And once someone's hired, most companies go quiet until something goes wrong. By then it's awkward for everyone.
Approach
Candidates go through AI voice interviews early, real conversations, not a form. Hiring managers see scores across communication, collaboration, and values fit, not just a gut feeling after one Zoom call.
The pipeline tracks people from first application through offer and into the first 90 days. Mismatches show up earlier, when you can still do something about it.
I wired up Pipecat and Daily for live voice, Python for evaluation logic, and the web app on Next.js and Hono. Getting all of that to work together in production was most of the job.
The Solution
Hiring managers see where each candidate is: applied, AI screening, team review, final interview, offer, with fit scores at every step. They can listen to transcripts, flag someone for a human review, or move them forward.
Candidates talk to an AI interviewer that asks real questions and evaluates how they respond. Not a chatbot reading a script. An actual conversation with signals a human can look at afterward.
After hire, the product tracks 90-day check-ins so teams notice when someone's struggling early, not at the annual review.
Key Challenges
-
live AI voice that doesn't lag, doesn't break when the connection drops, and produces transcripts managers can actually read
-
fit scores that mean something, specific enough that a hiring manager would trust them before making an offer
-
Python AI services, voice APIs, and the web app all talking to each other without fragile handoffs
-
making automation feel helpful, not like candidates are talking to a robot nobody will ever review
The Outcome
Teams get a clearer picture of fit before they extend an offer, not just a polished CV and a good first impression.
It's in production. Real hiring runs through it, from first screening to post-hire check-ins.
Context
Built at brigit.dev. I wasn't handed one feature on someone else's product. I built the thing.
The voice interview layer was the hard part. Pipeline UI and scoring are normal product work. Making a live AI conversation feel okay to sit through? That's where most of the risk was.