UNCENSORED AI

Inspiration

The project was born from the need for sovereign intelligence. Traditional AI research tools are often centralized, heavily filtered, and prone to data logging. The goal was to create a "Perplexity-style" experience that is completely uncensored, private, and runs on decentralized infrastructure, giving power back to investigative researchers and analysts.

What it does

Deep Agent is a deep investigation platform that goes beyond simple chat. It features three primary modes:

• Search: Quick, real-time web results with citations.
• Research: A multi-step pipeline that plans and executes parallel searches to synthesize comprehensive reports.
• Insight (Graph-RAG): The core innovation. It ingests documents and web data into a Neo4j Knowledge Graph, mapping relationships between entities like people, organizations, and dates to find hidden patterns.

How I built it

The app uses a "Sovereign Stack":

• AI Engine: Powered by Venice AI for LLM reasoning (deepseek-v3.2), web search (llama-3.3-70b), and vision (venice-v3-vision).
• Brain: Neo4j serves as the persistent memory, handling vector search and relationship mapping.
• Backend: Built with FastAPI and LangGraph to manage complex agentic workflows.
• Deployment: Configured for the Akash Network, a decentralized cloud, ensuring the platform remains permissionless.

Challenges I ran into

• Parallelization: Orchestrating multiple concurrent web searches and scrapers while maintaining a clean state in the agent's memory.
• Graph Mapping: Automatically extracting entities and creating meaningful RELATED_TO links in Neo4j from unstructured web text without generating "hallucinated" connections.
• Decentralized Deployment: Tuning the Akash SDL configuration to handle persistent storage and co-located services like Redis.

Accomplishments that I'm proud of

• Graph-RAG Integration: Successfully moving from flat text retrieval to a relational model where the AI can "see" the network of a case.
• Zero-Big-Tech Stack: Building a high-performance research tool without relying on OpenAI, Google, or AWS.
• Speed: Achieving high-intensity research reports through the use of ThreadPoolExecutors for parallel data gathering.

What I learned

• Agentic Workflows: The power of LangGraph for managing long-running research tasks that require decision-making at each step.
• Vector + Graph: Why vector search alone isn't enough for investigations; you need the explicit relationships that only a knowledge graph provides.
• Sovereign Infrastructure: The nuances of deploying to decentralized clouds like Akash compared to traditional VPS providers.

What's next for UNCENSORED AI

• Enhanced Memory: Wiring in the Redis and Supabase scaffolds for live case notifications and long-term user session storage.
• Collaborative Investigation: Allowing multiple researchers to interact with the same Knowledge Graph in real-time.
• Local-First Option: Providing an even easier path for users to run the entire stack on their own local hardware.

Built With

• akash
• fastapi
• pythoon
• supabase
• venice