AnveshakAI: A Decentralized AI

This project is building a decentralized AI-powered knowledge assistant that runs entirely on the Internet Computer Protocol (ICP). It's designed to be a personalized AI that becomes an expert in any domain you provide, while ensuring data privacy and decentralization through blockchain technology.

🛠️ Our Team

This project was crafted with immense dedication and hard work by our talented team:

• Aditya Dahuja (@Adhityadahuja)
• Bhavya Jain (@bhavyajain009)

We poured our hearts into building this innovative solution, and we're excited to share it with the world!

Visual Overview

flowchart TD
    A1["Frontend Interface (Next.js)"]
    A2["User Authentication & Identity (Internet Identity)"]
    A3["Canister Backend (Rust)"]
    A4["AI Engine (Google Gemini API)"]
    A5["Vector Search & Embedding System"]
    A6["Document Upload & Management"]
    A7["Streaming Response Module"]
    A8["DFX Deployment & Local Replica"]
    A9["Secure API Key Storage"]

    A1 -- "Authenticates via" --> A2
    A1 -- "Interacts with" --> A3
    A1 -- "Uploads documents to" --> A6
    A1 -- "Receives streaming output from" --> A7
    A1 -- "Visualizes data from" --> A5
    A3 -- "Handles logic for" --> A5
    A3 -- "Queries AI responses via" --> A4
    A3 -- "Stores & secures key in" --> A9
    A3 -- "Processes files from" --> A6
    A3 -- "Streams responses to" --> A7
    A8 -- "Builds & deploys" --> A1
    A8 -- "Deploys canisters" --> A3

Loading

🚀 Quick Start

Prerequisites

Before setting up the Anveshak AI, ensure you have the following installed:

• Node.js (v18 or higher) - Download here
• Rust (latest stable) - Install via rustup
• DFX (Internet Computer SDK) - Installation guide
• Git - Download here

Installation

1. Clone the repository

   git clone https://github.com/ChinmayBhattt/AnveshakAI-Decentralized-AI-Assistant.git
   cd AnveshakAI-Decentralized-AI-Assistant

2. Install dependencies

   # Install candid-extractor (required for deployment)
   cargo install candid-extractor

3. Start the local Internet Computer replica

   dfx start --background

4. Deploy the canisters

   dfx deploy

🔑 API Key Configuration

The Anveshak AI uses Google's Gemini API for AI responses. You'll need to configure an API key:

Step 1: Get a Google Gemini API Key

1. Visit Google AI Studio
2. Sign in with your Google account
3. Click "Create API Key"
4. Copy the generated API key

Step 2: Set the API Key in Your Canister

After successful deployment, configure the API key using the DFX command line:

dfx canister call backend set_api_key '("YOUR_GEMINI_API_KEY_HERE")'

Replace YOUR_GEMINI_API_KEY_HERE with your actual Google Gemini API key.

Important Security Notes:

• Only canister controllers can set the API key
• The API key is stored securely in the canister's memory
• Never share your API key publicly or commit it to version control

Step 3: Test the Setup

Verify everything is working by testing the backend directly:

dfx canister call backend prompt '("Hello, how are you?")'

You should receive a response like:

(
  variant {
    Ok = "I am doing well, thank you for asking! How are you today?\n"
  },
)

🌐 Access Your Assistant

Once deployed, you can access your Anveshak AI through:

• Local Development: http://localhost:4943/?canisterId=YOUR_FRONTEND_CANISTER_ID
• IC Network: https://YOUR_FRONTEND_CANISTER_ID.ic0.app

The canister IDs will be displayed after successful deployment.

🛠️ Development Commands

# Start local replica
dfx start

# Deploy all canisters
dfx deploy

# Deploy only backend
dfx deploy backend

# Deploy only frontend
dfx deploy frontend

# Check canister status
dfx canister status backend
dfx canister status frontend

# View canister logs
dfx canister logs backend

🔧 Troubleshooting

Common Issues

1. candid-extractor: not found

   cargo install candid-extractor

2. API key not set error

   • Ensure you've set the API key using the set_api_key function
   • Verify you're calling from a controller account

3. Frontend not receiving responses

   • Check that the backend is deployed and running
   • Verify the API key is properly configured
   • Check browser console for any JavaScript errors

4. DFX connection issues

   dfx ping local
   # If this fails, restart the replica:
   dfx stop
   dfx start --background

Core AI Capabilities

Intelligent Document Processing

• Multi-Format Support: Ingests and processes various file types, including text, PDFs, code files, and images.
• Automated Content Analysis: Automatically extracts, chunks, and creates embeddings for all uploaded content.
• Semantic Understanding: Leverages BERT-like transformer models for a deep semantic understanding of the text.
• OCR for Images: Includes Optical Character Recognition (OCR) to extract text from image-based documents.

Advanced Query System

• Natural Language Interface: A user-friendly, chat-like interface for asking questions in plain English.
• Vector Similarity Search: Retrieves the most relevant information using a high-performance vector similarity search.
• Coherent Responses with Citations: Generates clear, coherent answers and provides source citations for verification.
• Confidence Scoring: Each answer is accompanied by a confidence score to indicate the reliability of the information.

Adaptive Learning

• Instant Expertise: Becomes an expert in any uploaded domain within seconds.
• Cross-Document Analysis: Cross-references information from multiple documents to provide comprehensive answers.
• Contextual Conversations: Maintains context throughout a conversation for a more natural and intelligent interaction.

Technical Infrastructure

Backend (Rust)

• Efficient Vector Search: Implements the HNSW (Hierarchical Navigable Small World) algorithm for efficient vector search across over 1 million vectors.
• Contextual Understanding: Utilizes attention mechanisms to better understand the context of queries.
• Optimized Memory Management: Designed to handle over 10,000 documents per user with optimized memory usage.
• High-Speed Processing: Achieves sub-100ms processing time per document chunk.

Frontend (Next.js)

• Real-Time File Uploads: A drag-and-drop interface for file uploads with real-time processing status.
• Document Management: Features for previewing and categorizing uploaded documents.
• Analytics Dashboard: Provides usage metrics and other relevant analytics.
• Streaming Responses: Delivers responses as they are generated for a more interactive experience, with an option to export the conversation.

Unique Value Propositions

True Decentralization

• ICP-Powered: Runs entirely on the ICP blockchain with no reliance on external, centralized services.
• User-Owned Data: Ensures that users own their data, with built-in encryption for enhanced privacy.
• Censorship-Resistant: The decentralized architecture makes the assistant resistant to censorship.

Performance & Cost

• Affordable Model: A pay-once-to-upload model with unlimited querying.
• Fast Response Times: Delivers sub-2-second response times, even for complex queries.
• High Concurrency: Supports over 100 concurrent users.

Practical Use Cases

The Anveshak AI is designed to be versatile and can be applied to a wide range of scenarios:

• Medical: Upload medical textbooks and research papers to get diagnostic assistance.
• Legal: Upload legal documents and contracts to receive in-depth analysis and summaries.
• Development: Upload codebases to get explanations, identify potential issues, and receive suggestions for improvements.
• Cross-Domain: Handle a mix of document types for complex queries that span multiple fields of knowledge.

AnveshakAI-Decentralized-AI