A 60-minute lecture, processed to study materials in under four minutes.

University students spend 3–5 hours per lecture on work that adds no understanding — transcribing recordings by hand, extracting key concepts, building flashcard decks, writing practice questions. intelliQ wanted to automate the entire pipeline, but building it properly meant solving hard infrastructure problems: no cloud transcription vendor was fast enough, private enough, or affordable at the scale they needed.

Student audio is sensitive. Sending lecture recordings to a third-party transcription API was a non-starter — universities had data governance policies, and students deserved better. The platform needed a transcription engine that ran entirely within their own infrastructure, handled accented West African English accurately, and returned a usable transcript in minutes, not hours.

Transcription alone was not enough. The platform had to turn a raw transcript into a full study package — structured notes, a spaced-repetition flashcard deck, a graded quiz, and an AI tutor that could answer follow-up questions about what was actually said in the lecture, not just what the model knew from training.

A 60-minute lecture becomes a clean transcript in 2.5 minutes. Students upload anything — audio, video, PDF, Word document, YouTube link — and the system handles it without a cloud transcription service touching their data. NVIDIA Parakeet runs on a dedicated GPU in AWS; the audio stays there, the transcript comes back fast.

From that transcript, the platform generates structured notes organised by concept, a flashcard deck of 15–20 cards ready for spaced review, and a graded quiz where Claude evaluates each answer as correct, partial, or wrong. The AI tutor's responses are grounded in the student's own lecture — every answer cites the specific section that supports it, not general training data.

Students watch their lecture process live: uploading → transcribing → structuring → done, with partial transcripts appearing before the full lecture finishes. A real university lecture completed from job start to done in 3 minutes 58 seconds, confirmed in production Redis logs. Nothing falls through — failed jobs retry automatically, and a dead-letter queue preserves any that exhaust retries.

When a student asks a visual question — 'what does photosynthesis look like?', 'show me the circuit' — the tutor finds and displays relevant diagrams automatically. A keyword heuristic keeps this quiet for questions that don't need visuals. Each image lookup costs one credit from the student's balance, creating a natural rate limit without hard blocks.