← all projects

$ cat tldr.md

The agent runs the same voice pre-screen for every candidate, then hands the recruiter a transcript, a per-category scorecard, a strengths-and-weaknesses summary, and two integrity signals — a red-dot timeline of when other voices were detected, and a plagiarism/AI flag on every answer — all before anyone re-listens to the call.

$ cat problems.md

Why this needed building

$ ls features/ -la

What it does

Fixed-Script Voice Interviews

TTS asks the job's pre-set question list over a live call; the agent listens for each response before moving to the next question.

Live Answer Capture

After every answer, the agent calls a tool that stores it as a discrete record — the input every scoring step actually reads.

Post-Call Diarization

Once the call ends, the full recording is transcribed and speaker-diarized in one pass — not observed live.

Cheating Indicator Timeline

Diarization flags are plotted as red dots on the interview timeline — exactly when, and how often, another voice was heard.

Per-Answer Plagiarism / AI Check

Each stored answer is checked against a web search and an AI-generated-text classifier, independent of the transcript.

Category-Based Scoring

Each stored answer is graded against the job description and its target category — not one blunt overall number.

Recruiter-Ready Report

The post-call transcript for reading, per-category scores, a strengths/weaknesses summary, and every integrity flag — one document.

$ map pipeline/ --interactive

How an interview becomes a report

The technical pipeline from call to scorecard. Click a stage to see how it works.

recruitment-agent — architecture.svg
SETUP + LIVE CALL
LIVE CALL / POST-CALL
PER-ANSWER ANALYSIS
SYNTHESIS
REPORT

hover to trace · click a stage for the technical detail