בעיה
- Early-stage startup teams drown in scattered docs — SOPs, PDFs, Notion exports, onboarding guides
- New hires and busy founders waste time searching instead of executing
- Existing tools require complex setup or expensive enterprise contracts
מוצר עצמאי
Recall
AI knowledge assistant — ask your docs
Recall הוא עוזר ידע מבוסס AI שמאפשר לצוותים לשאול שאלות על המסמכים שלהם ולקבל תשובות עם ציטוטים מדויקים.
TypeScriptNext.js (App Router)OpenAI GPT-4oOpenAI Embeddings (text-embedding-3-small)Supabase (pgvector)Prisma 7
ציר זמן
Built as an independent POC
גודל צוות
Solo build
תפקיד
Founder / Full-stack Engineer
פתרון
- Built a RAG-based knowledge assistant that ingests any document format in under 2 minutes
- Implemented a chunking and embedding pipeline using OpenAI text-embedding-3-small and pgvector
- Designed a multi-turn chat interface that maintains session context and cites sources per answer
- Kept the stack lean — Next.js, Supabase, Prisma — so any startup can self-host or extend it
השפעה
- Any team can go from a pile of docs to a fully queryable assistant in under 2 minutes
- Demonstrates a production-grade RAG architecture with real citation traceability
- Provides a reusable foundation for internal copilots, support assistants, and onboarding tools
למה בניתי את זה
I built Recall to show how a lean RAG pipeline can replace hours of document hunting for startup teams — and to demonstrate the full architecture behind AI knowledge products, from ingestion to cited answers.
מה הייתי משפר בהמשך
- Add persistent conversation history across sessions with a database-backed chat store
- Introduce workspace-level auth so multiple teams can maintain separate knowledge bases
- Add streaming responses for a faster perceived UX on longer answers
- Support Notion, Google Drive, and Confluence as direct ingestion sources
