How mit-submit/a2rchi works
Main Components
a) Services Layer:
• Chat Service (service_chat.py)
• Cleo Service (service_cleo.py)
• Mailbox Service (service_mailbox.py)
• Piazza Service (service_piazza.py)
• Uploader Service (service_uploader.py)
• Account Creation Service (service_create_account.py)
b) Core AI Components:
• LLM Integration (chains/models.py)
• Chain Management (chains/chain.py, chains/base.py)
• Prompt Management (chains/prompts.py)
c) Data Storage:
• Vector Store (both HTTP ChromaDB and local options)
• SQL Database (utils/sql.py)
d) Interfaces:
• Chat Application (interfaces/chat_app)
• Uploader Application (interfaces/uploader_app)
• Cleo Interface (interfaces/cleo.py)
e) Utility Layer:
• Configuration Management (utils/config_loader.py)
• Data Management (utils/data_manager.py)
• Email Integration (utils/imap.py, utils/mailbox.py)
• Embedding Utilities (utils/embedding_utils.py)
• Web Scraping (utils/scraper.py)
[Client Layer]
[Interface Layer]
[Service Layer]
[Core AI Layer]
[Data Layer]
Deployment Aspects
A separate deployment view showing:
• Docker containers (from deploy/dockerfiles)
• Service composition (from deploy/*-compose.yaml files)
• Different deployment environments (dev, prod, etc.)
Additional Details
• Configuration management flow
• Where environment variables are used
• Points of extensibility
• Feedback loop for AI training
• Nested components
• Multiple connection types
• Export to high-resolution formats
Remember to maintain a balance between detail and clarity, focusing on the key architectural decisions while avoiding overwhelming detail at the component level.
This system design should reflect the modular nature of A2rchi, emphasizing its ability to work with different LLMs and its flexible data ingestion capabilities through various services.