StudyAI (autocards.app)
A live study platform used by 20 people, turning PDFs and images into flashcards with spaced repetition and RAG chat.
- result
- A live study platform with 20 real users. Measured embedding p95 is approximately 2.2 seconds at around 50 chunks per second; that measures embeddings, not the full document-to-flashcard workflow.
- my contribution
- Built the AI flashcard workflow, asynchronous document processing, and searchable study workspaces.
- 2024–Present
- web
how it works
StudyAI (autocards.app) is a live study platform with 20 real users. It turns PDFs and images into flashcards with OpenAI, schedules reviews with SM-2 spaced repetition, and answers questions over each user's documents through RAG chat. Photos of handwritten notes go through OCR first.
A study workspace keeps decks, progress, and recent sessions in one place. Document ingestion runs asynchronously with Celery on Railway, with documents stored in Cloudflare R2. Measured embedding p95 was approximately 2.2 seconds at around 50 chunks per second; this measures the embedding stage, not the full document-to-flashcard workflow.
engineering details
- Shipped a live study platform with 20 real users, with AI flashcards, SM-2 review scheduling, and RAG chat over each user's documents.
- Runs asynchronous document ingestion with Redis and Celery on Railway, with documents stored in Cloudflare R2.
- Measured embedding p95 at approximately 2.2 seconds at around 50 chunks per second.
- Integrated Supabase and pgvector to support searchable study content and user workspaces.
in the product
Flashcards From Notes and PDFs
An LLM splits any text or PDF into structured question-and-answer cards.
Study Workspace
Decks, progress, and recent study sessions in one place.
Photos of Notes
Photograph handwritten or printed notes and OCR turns them into study material.