Built an AI interviewing platform with adaptive questioning, real-time avatars, and browser-based proctoring

A complete AI job-interviewing platform supporting one-way and two-way interviews in 20+ languages, with customizable AI avatars and comprehensive evaluation reports. Candidates reach the interview through a redirect from the admin dashboard, where questions are dynamically generated and read aloud via text-to-speech, with an AI-determined, question-specific response timer for each answer.

Interviews are adaptive rather than scripted: after every answer, the AI checks the candidate's response against a predefined set of interview criteria and dynamically generates targeted follow-up questions for anything not yet covered — ending the interview automatically once all criteria are satisfied. A post-interview scoring pipeline then passes the full question-answer transcript and criteria matrix to an LLM, producing a configurable score (out of 10, 100, or whatever scale a company uses) alongside a dynamic report with per-question feedback and per-criteria evaluation.

Alongside the core interview flow, the platform includes a mock-interview and skill-assessment module for candidate practice, and a browser-based proctoring system built from scratch on Google MediaPipe — running entirely client-side via WASM, with per-frame video analysis and no backend processing, to flag mobile phone usage, multiple people in frame, and multiple simultaneous voice sources. All flags surface directly in the final interview report.

I built the candidate-facing interview UI end-to-end, including the text-to-speech question delivery and the adaptive follow-up logic that checks answer coverage against interview criteria in real time.

I designed the post-interview scoring pipeline, integrated Azure AI's real-time avatar system for the interactive interview experience, and built the mock-interview and skill-assessment modules. Separately, I developed the client-side proctoring system from scratch using Google MediaPipe — no backend video processing, all analysis running per-frame in the browser via WASM.