Inspiration
The initial screening process is often the biggest bottleneck in hiring. Recruiters frequently deal with hundreds of applications for a single role, making it nearly impossible to conduct meaningful early-stage interviews with everyone. We wanted to build a tool that gives every candidate a chance to "speak" to the company while providing recruiters with immediate, actionable data to identify the best fits quickly.
What it does
Vetted facilitates automated AI voice interviews between a recruiter (represented by the AI) and an applicant. Once the call concludes, the system processes the transcript to extract specific metrics assigned by the recruiter such as technical knowledge, work ethic, and factuality. These metrics are then delivered in a structured format, allowing recruiters to sort and filter candidates based on performance rather than just a resume.
How we built it
The application is built using Python and JavaScript. To ensure the AI doesn't just provide "fluff" in its evaluation, we implemented strict response schemas using Pydantic and Zod. This forces the model to return numerical metrics in a clean JSON format, which can then be easily ingested into a database or spreadsheet.
Challenges we ran into
One of the biggest hurdles was managing API reliability. During peak times, LLM providers often return 503 UNAVAILABLE errors. We had to implement a robust exponential back off strategy to ensure that calls and transcript processing wouldn't fail during high-traffic periods. Additionally, prompt engineering for "factuality" metrics proved difficult ensuring the AI can distinguish between a confident-sounding answer and a correct one.
Accomplishments that we're proud of
I’m particularly proud of the data consistency. By using structured output, we transformed a subjective, conversational interview into an objective set of data points. Achieving a system that can reliably handle API retries without losing state or doubling up on calls was also a significant technical win.
What we learned
Building Vetted taught me a lot about the nuances of structured data extraction. I learned that even the most advanced AI models need "guardrails" (like schema enforcement) to be useful in a professional business context. I also gained a deeper understanding of asynchronous processing and how to handle distributed system failures gracefully.
What's next for Vetted
The next step is to build out a full-scale recruiter dashboard that visualizes these metrics across entire candidate pools. We could integrate with popular Applicant Tracking Systems (ATS) so that these "Vetted" scores can be pushed directly into a company’s existing workflow.
Log in or sign up for Devpost to join the conversation.