Lecsum

Turn lecture recordings and documents into study materials. Upload an audio file, PDF, Word doc, or image — get a transcript, summary, quiz, and flashcards powered by AI.

What it does

1. Upload an MP3, WAV, M4A, FLAC, PDF, DOCX, PPTX, or image
2. Audio → AWS Transcribe converts speech to text
3. Documents/images → Amazon Bedrock Claude Haiku extracts all text and describes diagrams via vision
4. AI generates a summary, quiz questions, and flashcards from the content
5. Study in the browser with an editorial-style study page, color-coded per course
6. Organize lectures into courses, drag to assign or trash

Tech stack

Layer	Tech
Frontend	Next.js 15, Tailwind CSS, TypeScript
Backend	AWS Lambda (Python 3.12), Next.js API routes
Transcription	AWS Transcribe
AI / Vision	Amazon Bedrock (Claude Haiku 4.5, Nova Pro)
Storage	S3 (3 buckets), DynamoDB
CI/CD	GitHub Actions
Deployment	Vercel

Architecture

Any file upload → S3 (raw-uploads)
        ↓ S3 trigger
Lambda: router
    ├── Audio → starts Transcribe job → DynamoDB status: transcribing
    │       ↓ Transcribe finishes
    │   S3 (transcripts-outbox) → S3 trigger
    │   Lambda: transcript-parser → extracts text → S3 (transcripts-text) → DynamoDB: done
    │
    └── PDF / Image → DynamoDB status: extracting
            ↓ browser calls /api/extract
        Next.js API → Bedrock Claude Haiku vision → S3 (transcripts-text) → DynamoDB: done

Dashboard polls DynamoDB every 10s → card updates live when done
Study page → /api/summary, /api/quiz, /api/flashcards → Bedrock Nova Pro
           → responses cached in S3 (.summary.json, .quiz.json, .flashcards.json)

Project structure

lecsum/
├── .github/
│   └── workflows/
│       └── deploy.yml              # CI/CD pipeline
├── lambdas/
│   ├── router/                     # S3 trigger — routes by file type
│   │   └── lambda_function.py
│   ├── transcribe_trigger/         # starts Transcribe job, writes DynamoDB record
│   │   └── lambda_function.py
│   └── transcript_parser/          # extracts text, updates DynamoDB to done
│       └── lambda_function.py
├── lecsum-web/                     # Next.js frontend
│   └── app/
│       ├── dashboard/              # main dashboard — courses, lecture cards, upload
│       ├── processing/             # background status (legacy, kept for direct access)
│       ├── study/                  # editorial study page with sidebar nav
│       └── api/
│           ├── upload-url/         # presigned S3 URL
│           ├── job-status/         # polls DynamoDB
│           ├── lectures/           # scan all jobs, PATCH course/displayName, DELETE
│           │   └── [uploadKey]/
│           ├── courses/            # GET/POST courses list (stored in DynamoDB)
│           ├── extract/            # Bedrock vision extraction for PDFs/images
│           ├── transcript/         # reads .txt from S3
│           ├── summary/            # Bedrock summary with S3 cache
│           ├── quiz/               # Bedrock quiz with S3 cache
│           └── flashcards/         # Bedrock flashcards with S3 cache
├── scripts/
│   └── deploy.sh                   # packages deps + deploys Lambdas per env
├── tests/
│   ├── unit/
│   └── integration/
├── requirements-dev.txt
└── pyproject.toml

S3 buckets

Bucket	Purpose
lectureai-raw-uploads-dev	incoming audio, PDF, image files
lecsum-transcripts-outbox	raw Transcribe JSON output
lecsum-transcripts-text	plain text transcripts + cached AI outputs

The lecsum-transcripts-text bucket stores:

• <key>.txt — extracted transcript
• <key>.summary.json — cached summary
• <key>.quiz.json — cached quiz
• <key>.flashcards.json — cached flashcards

DynamoDB

Table: lecsum-jobs — partition key: uploadKey

Field	Description
uploadKey	S3 key of the uploaded file
jobName	Transcribe job name (audio only)
status	transcribing | extracting | done | error
transcriptKey	S3 key of the output .txt file
fileName	original filename
displayName	user-renamed label
course	course ID (from lecsum-courses item)
createdAt	ISO timestamp

Courses are stored as a single item with uploadKey: "courses" and a data field containing the JSON array.

CI/CD

Every push to staging → tests → lint → deploys to staging Lambdas.
Every push to main → tests → lint → deploys to prod Lambdas.

Lambda naming convention: lecsum-<function>-<env>

Local development

Prerequisites

• Node.js 20+
• Python 3.12
• AWS account with Bedrock, Transcribe, S3, DynamoDB, Lambda access
• Bedrock model access: claude-haiku-4-5 and amazon.nova-pro-v1

Frontend

cd lecsum-web
npm install
cp .env.example .env.local   # fill in your AWS credentials
npm run dev

Environment variables

AWS_REGION=us-east-2
AWS_ACCESS_KEY_ID=...
AWS_SECRET_ACCESS_KEY=...
S3_UPLOAD_BUCKET=lectureai-raw-uploads-dev
S3_TRANSCRIPTS_BUCKET=lecsum-transcripts-text

Tests

pip install -r requirements-dev.txt
pytest tests/unit/ -v

Linting

ruff check lambdas/
ruff format lambdas/

Roadmap

• Audio upload with presigned S3 URL
• AWS Transcribe pipeline
• DynamoDB job tracking
• Real-time processing status (background, no redirect)
• Multi-file support — PDF, DOCX, PPTX, images via Bedrock vision
• Summary tab (Bedrock Nova Pro)
• Quiz tab with scoring
• Flashcard tab with spaced repetition UI
• S3 caching for AI outputs
• Dashboard with course organization
• Drag-to-assign courses, drag-to-trash
• Rename lectures and courses inline
• Animated bubble panel for course cards
• Dark / light theme toggle
• Editorial study page with course color theming
• Courses persisted in DynamoDB
• Auth (Clerk)
• DOCX / PPTX native support (currently: convert to PDF)
• Anki .apkg export
• Bookmark and PDF export on study page
• Mobile app (Expo)
• A2A agent orchestration (Bedrock AgentCore)
• MCP integrations (Notion, Google Calendar auto-tagging)