Skip to content

Latest commit

 

History

History
75 lines (52 loc) · 3.45 KB

TECHNICAL_ARCHITECTURE.md

File metadata and controls

75 lines (52 loc) · 3.45 KB

_# Ajatuskumppani - Technical Architecture

This document provides a detailed overview of the technical architecture of the Ajatuskumppani project.

System Overview

The system is designed as a modular, microservice-based architecture. The core components are:

  • AjatusUI: The user-facing application.
  • AjatusServer: The central backend service.
  • AjatusCore: The AI model and inference engine.
  • AjatusMemory: The long-term memory and user profile service.
  • AjatusNode: The decentralized compute nodes.
  • AjatusAgents: Autonomous agents for specific tasks.
  • AjatusToken: The blockchain-based token.

System Architecture

Component Details

AjatusServer (FastAPI)

  • Authentication: JWT-based authentication with access and refresh tokens.
  • API: RESTful API for all standard operations. WebSockets for real-time communication with AjatusUI.
  • Database: PostgreSQL with pgvector for structured data and vector embeddings. SQLAlchemy as the ORM.
  • Caching: Redis for caching frequently accessed data and for message queuing.
  • Task Orchestration: Celery for background tasks and for orchestrating the AjatusAgents.

AjatusCore (Python)

  • Model: Fine-tuned version of Mistral 7B Instruct.
  • Inference: vLLM for high-throughput serving on GPUs. Ollama as a fallback and for CPU-based inference.
  • API: A simple internal API for the AjatusServer to get model completions.

AjatusMemory (Python)

  • Embedding Generation: Uses sentence-transformers to generate embeddings for user interactions.
  • Vector Storage: pgvector for storing and querying embeddings. FAISS for in-memory indexing and faster retrieval.
  • User Profile: A service that continuously updates a user's profile based on their interactions and retrieved memories.

AjatusNode (Python/Docker)

  • Compute Environment: Docker containers for securely and isolating AI tasks.
  • Proof-of-Contribution: A custom algorithm that measures uptime, performance, and task validation to reward nodes.
  • Networking: A P2P network for communication between nodes.

AjatusUI (React Native)

  • State Management: Zustand for simple and scalable state management.
  • API Communication: Axios for REST API calls and the native WebSocket API for real-time updates.
  • Styling: Tailwind CSS for a consistent and modern look and feel.

AjatusToken (Solidity)

  • Contract: An ERC-20 compliant token contract.
  • Staking: A staking contract that allows users to lock up their tokens and earn rewards.
  • Governance: A governance contract that allows token holders to vote on proposals.

Data Flow

  1. A user interacts with the AjatusUI.
  2. The UI sends a request to the AjatusServer.
  3. The server authenticates the user and processes the request.
  4. If the request requires AI completion, the server sends a prompt to the AjatusCore.
  5. The AjatusCore returns the completion to the server.
  6. The server stores the interaction and its embedding in the AjatusMemory.
  7. The server returns the response to the UI.

Security

  • All communication between services is encrypted with TLS.
  • User data is encrypted at rest.
  • The AjatusNode uses Docker containers to isolate tasks and prevent malicious code from accessing the host system.
  • Regular security audits will be performed.