AI Notebook
Collaborative AI Research & Knowledge Platform
Enterprised SurfSense (open-source AI notebook) into a self-hosted research platform where teams connect 25+ data sources, ask natural language questions with cited answers, collaborate in real-time, and generate podcast summaries. Private alternative to Perplexity and NotebookLM that keeps data on-premise.
Teams needed a private, self-hosted research tool that could connect to internal knowledge sources like Google Drive, Slack, Jira, and Confluence while keeping sensitive data within the organisation's infrastructure.
Lead engineer. 4 ECS services (backend, workers, scheduler, sync), cross-account model access via STS AssumeRole, infrastructure across DEV/STG/PRD.
Highlights
Hierarchical 2-tier RAG
Hybrid semantic search (pgvector) and BM25 full-text search merged with Reciprocal Rank Fusion. Multiple rerankers available for result quality tuning
Deep agent orchestration
LangGraph-based agents with sub-agent support. 8 registered tools: knowledge base search, image generation, podcast creation, web scraping, MCP client, link preview, user memory, and document search
25+ data source connectors
Google Drive, Slack, Teams, Jira, Notion, GitHub, Confluence, Discord, Airtable, Linear, and more. Celery workers handle async document ingestion and vector indexing
Real-time team collaboration
Electric SQL enables Firebase-like sync on PostgreSQL. Role-based access control (Owner/Admin/Editor/Viewer), comment threads on messages, and shared knowledge spaces