Project 04 · Hackathon build
Knowledge Twin
Knowledge Twin - AI Powered Chat Bot
Documents, images, video and memory in one console.
Knowledge Twin is a powerful RAG-driven AI built to act like a world-class knowledge expert—able to read documents, understand images and videos, and remember conversations with exceptional accuracy.
- Type
- Hackathon build
- Built at
- Dev Heat Hackathon
- Core stack
- React · Vite · Tailwind CSS
- Live at
- knowledgetwin.netlify.app
Overview
What it is
Knowledge Twin is a retrieval-augmented generation (RAG) assistant. It reads documents, understands images, watches short videos and remembers the conversation, then answers from the material you gave it.
Uploaded documents are split into chunks and embedded as vectors, so a question finds the right context even when the wording differs. Answers are grounded in your own data rather than guessed.
Architecture
How it’s built
- 01React frontendChat, documents, media and analytics
- 02Express APIUploads, queries and dashboard routes
- 03RAG + memoryChunking, retrieval and chat history
- 04Gemini + MongoDBGeneration and vector storage
Features
What it does
Multimodal
- Image understandingExplains photos and diagrams, including text inside images.
- Video understandingSummarises MP4 clips and answers questions about them.
- Media in chatVideo player and image previews inside the conversation.
Retrieval (RAG)
- Semantic searchVector search over document chunks finds the exact context.
- PDF summariesDistils long PDFs into short, high-impact bullet points.
- Grounded answersResponses are built from your uploaded data.
Memory and resilience
- Conversation memoryKeeps recent messages so follow-up questions just work.
- Automatic retriesBacks off and retries when the AI API is rate-limited.
- Model fallbackDiscovers available Gemini models and falls back through them.
Tech stack
Built with
Frontend
- React
- Vite
- Tailwind CSS
- Framer Motion
- Recharts
- React Markdown
- Axios
Backend
- Node.js
- Express
- MongoDB
- Mongoose
- Multer
- pdf-parse
AI
- Google Gemini API
- Vector embeddings
Screens
A closer look
Built at a hackathon
Dev Heat Hackathon
Top 15among 200+ teams
- Organiser
- IIIT Surat
- Platform
- Unstop
- Date
- Feb 2026
- Format
- Team event
- Team
- 4 members
- Focus
- AI · Web
- Our build
- Knowledge Twin AI-Chat Bot