Bhishaj/Anvesha-AI
Anvesha AI
Sovereign Intelligence Engine for Bharat
Anvesha AI is a sovereign Indian intelligence platform engineered to transform raw search data into strategic, citation-rich insights. It combines cutting-edge Indic LLMs with a privacy-first architecture to empower researchers, policymakers, and citizens with deep, actionable intelligence.
Core Technology (The Sutra Pipeline)
Router: Sarvam 30B
The Sarvam 30B model acts as an intelligent query optimizer, decomposing complex natural-language questions into 3-5 precise, parallel search queries. It focuses on Indian context and prioritizes government sources (.gov.in) to ensure strategic relevance.
Synthesizer: Sarvam 105B
The powerful Sarvam 105B model synthesizes aggregated search results into a high-density, multi-section 'Sutra' report. It extracts every metric—dates, version numbers, percentages—and structures them into detailed narratives with inline citations, dynamic follow-up questions, and explicit focus on Indian policy impacts.
Verification: Prioritized .gov.in Indexing
A custom-configured SearxNG instance, tuned for Indian relevance and stripped of resource-heavy engines, serves as the verification layer. It ensures that authoritative government sources are indexed first and surfaced prominently in every intelligence report.
Multimodal Capabilities
- Saaras V3 (Speech-to-Text): Enables highly accurate Hindi and English voice input, allowing users to speak their research queries naturally.
- Bulbul V3 (Text-to-Speech): Converts synthesized intelligence reports into natural-sounding Indian voice audio, supporting eyes-free consumption and accessibility.
UX Features
- Interactive Source Panel: A slide-out panel displaying all cited sources with domain badges, GOV.IN indicators, and clickable URLs for deep verification.
- Dynamic Follow-ups: Context-aware suggested questions generated by the synthesizer, enabling iterative, Perplexity-style exploration.
- Search Cancellation: Real-time abort capability via an intuitive Stop button, giving users full control over long-running LLM inferences.
- Collapsible Sidebar: A responsive navigation panel that can be hidden to maximize focus on the synthesis view.
Technical Architecture
Anvesha AI runs as a unified monolith container managed by Supervisord. The stack comprises:
- Frontend: Next.js (Standalone mode) with Tailwind Typography and responsive sliding panels.
- Backend: FastAPI application served by Gunicorn with Uvicorn workers, orchestrating the Sarvam API calls.
- Search Engine: SearxNG instance operating on port 8888, providing the verified search foundation. All services communicate over localhost and are orchestrated to start together via a single
docker compose upcommand.
Quick Start
To run Anvesha AI locally for development:
- Prerequisites: Docker Desktop installed and running, Git installed.
- Clone the repository:
git clone https://github.com/Bhishaj9/Anvesha-AI.git
cd Anvesha-AI- Configure Environment: Copy the template and add your Sarvam AI API key:
cp backend/.env.template backend/.env
# Edit backend/.env and set SARVAM_API_KEY=your_key_here- Build and Launch:
docker compose up --build- Access the Application: Open your browser to http://localhost:3000
The backend will be available at http://localhost:8000 and SearxNG at http://localhost:8080.
Legal & Contribution
We welcome community contributions to improve the sovereign search experience! If you're interested in refining the Sutra Pipeline, enhancing multimodal capabilities, or fixing bugs, feel free to open a Pull Request.
Ownership & License Statement: Anvesha AI is a proprietary project of Bhishaj Technologies. While the code is visible for contribution, all publishing and distribution rights belong exclusively to Gaurav. Please see the LICENSE file for more details.
