AnveshakAI

Inspiration Traditional AI assistants rely heavily on centralized infrastructure, raising concerns around data privacy, ownership, and transparency. We wanted to build a system where users can fully control their data while still leveraging powerful AI capabilities. AnveshakAI was inspired by the idea of combining decentralized technologies with intelligent systems to create a trustless, private, and domain-adaptive AI assistant that anyone can use securely. What it does AnveshakAI is a decentralized AI-powered knowledge assistant built on the Internet Computer Protocol (ICP). It allows users to: Upload documents (PDFs, text, code, images) Automatically process and understand content Ask natural language questions Receive context-aware answers with citations and confidence scores The system becomes an instant expert in any domain based on the data provided by the user. How we built it We designed AnveshakAI using a modern decentralized + AI stack: Frontend: Next.js for an interactive UI with real-time streaming responses Backend: Rust-based ICP canisters for secure and efficient logic execution Authentication: Internet Identity for decentralized user authentication AI Engine: Google Gemini API for generating intelligent responses Vector Search: HNSW-based embedding system for semantic retrieval Document Processing: Automated chunking, embedding, and OCR for images Deployment: DFX SDK for deploying on ICP The architecture ensures seamless interaction between decentralized storage, AI inference, and real-time user experience. Challenges we ran into Integrating AI APIs with decentralized canisters securely Managing API key storage without compromising security Implementing efficient vector search within blockchain constraints Handling large document processing with optimized memory usage Ensing smooth real-time streaming responses from backend to frontend Accomplishments that we're proud of Built a fully functional decentralized AI assistant on ICP Achieved sub-2 second response times Implemented vector search for over 1M embeddings Enabled multi-format document understanding (including OCR) Delivered a clean, real-time user experience with streaming outputs

What we learned How to combine Web3 + AI effectively in a production-like system Deep understanding of ICP architecture and canister development Practical implementation of vector databases and semantic search Importance of secure key management in decentralized systems Designing systems for scalability, latency, and usability simultaneously What's next for AnveshakAI Fully on-chain AI inference (reducing dependency on external APIs) Enhanced fine-tuning for domain-specific models Multi-user collaboration on shared knowledge bases Mobile-friendly interface and broader accessibility Integration with more decentralized storage solutions

Built With

• icp
• mern