Overview
An agentic knowledge base that lets you upload documents, ask questions, and get cited answers — powered by hybrid search, reranking, and a LangGraph agent that decides when to search your docs vs the web.
(Preview)
See it in motion
Demo
Features
- (001)Many sourcesPDFs, websites, YouTube videos, images, Word docs, PowerPoints, CSVs and markdown files
- (002)Natural language or voice questions
- (003)Exact source citations on every answer
- (004)Agent searches your documents first, falls back to web search if needed
- (005)Detects contradictions between sources
- (006)Flags knowledge gaps and suggests resources
- (007)Generates quizzes from your uploaded content
- (008)Builds learning roadmaps from a single goal
Tech stack
- Python
- FastAPI
- LangGraph
- ChromaDB
- BackendPython, FastAPI, LangChain, LangGraph
- Vector DBChromaDB
- RetrievalHybrid search (vector + BM25), BGE cross-encoder reranking, multi-query RAG
- LLMGemini (SaaS) / Ollama + Llama 3 (self-hosted / private)
- DatabasePostgreSQL
- AuthJWT with httpOnly cookies
- FrontendReact (in progress)
- DevOpsDocker, GitHub Actions
Architecture
- User question
- Multi-query generation (3 query variants)
- Hybrid search — vector search + BM25 in parallel
- BGE reranker — top 20 → top 5
- LangGraph agent — docs enough? or search web too?
- Answer with citations + contradiction flags
Deployment modes
- SaaSGemini API — standard privacy
- Self-hostedOllama + Llama 3 — 100% private, data never leaves your server
- EnterpriseAzure OpenAI — private cloud
Status
- BackendIn progress
- FrontendComing soon
- DockerComing soon