Skip to content

Portfolio

Personal projectResearch projectAI system

TIPS

A six-stage Python pipeline that analyses recorded interviews across speech, audio and video to produce scores that show how a candidate's answers change over time.

The challenge

Interview judgements are subjective and hard to compare, and a single overall score hides how a candidate's performance changes during a conversation.

The approach

Split analysis into six stages: synchronise the timeline, extract audio and video features, segment the interview, analyse behaviour, score answers semantically with an LLM, then aggregate the results.

What we did

Abrar designed and built the pipeline and its analytics dashboard. The related paper, Temporal Stability-Aware Multi-LLM Consensus for Candidate Evaluation, dates from 2026.

The solution

A batch system that turns a recorded interview into a timeline of behavioural and semantic scores, shown in an interactive dashboard with question-by-question review.

Key features

  • Whisper speech-to-text and automatic question and answer pairing
  • Audio and video feature extraction with OpenCV and MediaPipe
  • LLM-based semantic scoring of answers
  • Time-based scoring that tracks change across the interview
  • Interactive analytics dashboard

Where it stands

It is a research system for recorded interviews, not a hiring product in use by employers. The screenshot shows sample analysis output.

Have a project in mind?

Tell us what you need. The first 30-minute call is free, and we'll talk it through with you and give an honest view of whether we are the right fit. We reply within 24 hours, usually sooner.

Not ready to talk yet? Build a project brief first