Overview
What this challenge is about.
You receive about 40 hours of de-identified clinician voice notes paired with corrected transcripts plus a medical-terminology lexicon (about 8,000 drug + procedure terms). Fine-tune Whisper (medium or large-v3) or a comparable open-source ASR model with domain-adaptive training. Measure WER overall and on a medical-term-only slice. Hit 30 percent relative WER reduction on the medical-term slice without hurting overall WER. Deliver the fine-tuned checkpoint, training notebook, evaluation report, and a 2-page ship/no-ship memo.
The Brief
What you'll do, and what you'll demonstrate.
Fine-tune an open-source ASR model to cut medical-term WER by 30 percent without hurting overall WER.
Earning criteria — what you'll demonstrate
- Fine-tune a modern ASR model on a domain corpus
- Evaluate ASR with overall and sliced WER
- Avoid overfitting the medical-term slice at the cost of general WER
- Communicate a model-ship decision to product leadership
Program Fit
Where this fits in your program.
Sharpens the same skills your degree expects you to demonstrate.
Skills
Skills you'll demonstrate.
Each one shows up on your verified credential.
Careers
Roles this prepares you for.
Real titles. Real skill bridges. Pick the one closest to your trajectory.
Career paths this builds toward
Canonical rolesNLP Engineer
Domain-adapting an ASR model and shipping a quantitative ship/no-ship memo is the day-to-day work of NLP engineers at speech-heavy SaaS companies.
This challenge sharpens
- asr
- speech-recognition
- fine-tuning
Machine Learning Engineer
Fine-tuning with rigorous sliced evaluation translates directly to the MLE's broader work on production-quality models.
This challenge sharpens
- domain-adaptation
- wer-evaluation
- fine-tuning
Applied AI Scientist
Connecting model improvements to a product metric (coder cleanup minutes) is the applied-AI scientist's job in vertical-AI startups.
This challenge sharpens
- asr
- wer-evaluation
- domain-adaptation