The Problem
Turns long recorded lectures into short, high-value study clips by pairing transcript analysis with visual concept matching.
MULTIMODAL AI · EDTECH
StudySlice
StudySlice cuts hours of lecture recordings down to the moments that actually matter. AWS Transcribe converts the audio to text, AWS Bedrock analyzes the transcript to separate real concept explanations from filler, and CLIP-based visual matching pairs each identified concept with the right video frame — all orchestrated through a FastAPI pipeline built around a system-design-first, ingestion/analysis/clip-generation architecture.
GitHub · soonLive demo · soon
Stack
AI + ML
CLIP
Backend + Data
PythonFastAPI
Frontend
Next.jsTypeScript
Infrastructure
AWS BedrockAWS TranscribeAWS S3
What I Built
StudySlice cuts hours of lecture recordings down to the moments that actually matter. AWS Transcribe converts the audio to text, AWS Bedrock analyzes the transcript to separate real…
The Difference
AWS Bedrock analyzes lecture transcripts to flag high-value concept segments versus filler content
Numbers That Matter
8primary scale
3measured result
01system signal
E2Equality outcome
Engineering Notes ✎
Engineering Note 1
AWS Bedrock analyzes lecture transcripts to flag high-value concept segments versus filler content
Engineering Note 2
CLIP-based visual matching pairs identified concepts with the correct video frames
Engineering Note 3
System-design-first architecture separates ingestion, analysis, and clip-generation into independently scalable stages
Architecture Overview
Lecture video↓Content Pipeline
TranscriptionConcept AnalysisCLIP Matching
AWS S3↓Focused study clips