Interactive product sprint case study
From CV intake to evidence worth reviewing.
InterviewMate turns batch resume screening and first-screen interviews into one traceable recruiter workflow. This page explains the product decision, the AI boundaries and the evidence behind the MVP.
Role brief → CV ranking → interview → evidence → recruiter history
Temporary admin credentials are included in the private submission PDF.
Use the reviewer account from that PDF to explore the full workspace. The account is temporary and expires on 18 September 2026 at 00:00 UTC.
The five-day arc
Every phase leaves a usable decision behind.
Choose a phase to see the question it answered and the artifact it produced.
Find the evidence bottleneck
A recruiter needs a concise, job-specific record from CVs and first-screen answers. The sprint started with desk research and kept the problem statement explicit as a hypothesis.
The product loop
One workflow, with a clear handoff at every step.
Recruiters can stop after ATS ranking or continue through invitations. Candidates only see their own session; the workspace keeps the review record.
Role brief
Enter one job description for the entire CV batch.
Hands-on product brief
Walk the workflow before you read the report.
This small, self-contained walkthrough uses fictional data. Choose a step, inspect the decision it exposes and then open the live prototype when you want to try the real interaction.
Start with one role brief
The recruiter enters the job description once. It becomes the shared context for ATS ranking, interview prompts and the evidence rubric.
AI logic
Models assist the conversation. Rules protect the record.
The implementation separates speech, transcription and evaluation. A generated answer cannot become a saved assessment until its structure and candidate quotations pass validation.
Bounded context
Role brief, validated CV text and relevant GitHub context are limited before they reach the interview or evaluator.
Candidate control
Speech becomes an editable draft. The candidate chooses when to send it, and the draft clears after submission.
Evidence validation
Exact candidate quotes, turn references and competency levels are checked before the percentage is displayed.
Evaluation and limits
Readable scores, explicit uncertainty.
The score describes supported interview evidence under a rubric. It is not a hiring recommendation or a probability of success.
A normalized view of supported competency evidence, not a hiring probability.
How many configured competencies have eligible candidate evidence.
A deterministic CV/JD signal used to prioritize recruiter review.
Experience ratings for clarity, transcription, relevance and reliability.
Authored transcripts, ten fictional CVs and builder-led English/Indonesian checks.
They were useful for detecting behavior regressions and validating the workflow. They do not establish recruiter time savings, inter-rater reliability or model accuracy.
Measure manual versus assisted review before claiming productivity.
Use comparable synthetic cases, record corrections and independently inspect quotes and competency levels. Keep the gap visible in the submission.
Engineering handoff
Enough detail to build the next version without guessing.
The full PDF contains the source map, API contract, persistence boundaries, runbook and selected reports. The interactive view keeps the key decisions easy to scan.
What is deterministic
ATS matching, evidence percentage math, transcript quote validation, access boundaries and session recovery state.
What uses a model
Voice conversation, transcription and structured evidence interpretation. Provider output is validated before it can become a score.
What remains human
Candidate selection, evidence interpretation, hiring judgment, baseline measurement and approval of any future learning dataset.
What is deliberately deferred
Automatic rejection, covert monitoring, OCR, HRIS/email integration and fine-tuning on raw candidate data.
Continue the review
Open the prototype, then keep the evidence close.
Use the live demo for the experience and the PDF for the complete case study and supporting reports.