Build a lecture-audio to structured-notes tool
Speech-to-insight tools are common in modern workplaces. Build your own: record real audio (a class, an online lecture, or a family explanation — with consent), transcribe it with a free AI model, and have a large language model (LLM) turn the transcript into structured notes automatically.
Posted by The AI Internships
The work brief
- 01Get 2 real recordings with consent (10+ minutes total): a class, a study-group session, or a lecture you play and record.
- 02Build the pipeline: audio → transcription (Whisper locally/Colab, or Gemini’s audio input) → LLM pass that structures it into topics, key points, and action items.
- 03Compare output against the actual audio: mark errors caused by your accent, code-switching, or technical terms, along with structure mistakes.
- 04Fix the worst failure mode with prompt or code changes, and push everything to a public repo.
What you’ll produce
Submission standard
Submit the repo, a demo video of the pipeline running, one full before/after (raw transcript → structured notes), and the accuracy audit. Include your AI workflow: tools, best prompts, and what you changed from the AI’s output. Your repo must show the pipeline built across more than one commit, and the audit must cite errors found against the actual audio.
Public repo with the pipeline code
Public repo with the pipeline code — commit history should show transcription and the LLM structuring pass wired across multiple commits, not one paste. A single dump commit fails the authenticity gate.
Public linkRequiredPublic demo video of the pipeline processing real audio
Public linkRequiredOne real example: raw transcript excerpt → your structured notes output
Written responseRequiredThe audit: transcription errors found against the real audio, structure mistakes, and the fix you shipped
Written responseRequired
You’ll complete these inside your private workspace.
How your work is evaluated
The passing benchmark is 70/100.
Working pipeline
29%Real audio goes in, structured notes come out, per the demo.
Accuracy audit
29%Errors were found by checking against the actual audio — not assumed.
Notes quality
14%The structured notes would genuinely help a student who missed the class.
Real pipeline, real audio audit
29%Fails if the demo link is dead, the repo is a single dump, or the audit shows no errors checked against the real audio (accent/code-switching/technical-term mistakes a clean paste wouldn't have).
How we grade your AI usage
Using AI is the point — it’s the skill this certificate proves. You’ll answer three short questions about how you used it: what you asked, what was wrong with its first answer, and what you changed. Specific, honest answers score high. “I pasted the brief and submitted the answer” scores near zero.