GenAI Blueprint

This project has several goals:

• Support GenAI training, to illustrate key concepts, programming patterns and techniques
• Implement the Factory pattern for LLM, Embeddings models, Vector Store, Runnables
• Act as a repository of well integrated reusable components
• Act as a template for cloud-deployable, highly configurable AI Agents
• Propose ready to use agentic GenAI demos, demonstrating notably:
  • ReAct Agents to connect different data sources
  • Tool calling agent calling an API
  • Semantic and Hybrid search
  • Researcher agents
  • MCP
  • CrewAI

It's based mainly on the LangChain ecosystem, and integrate many other nice solutions.

Core dependencies:

• LangChain, LangGraph, LangServe, ...
• OmegaConf: configuration management
• Streamlit: User Interface (web)
• Typer: Command Line Interface
• FastAPI: REST APIs
• Pydantic for .. a lot

Extra components used:

• SmallAgents : Agent Framework
• GPR Researcher : Deep Internet Search
• MCP Adapt : Model Contect Protocol
• tavily-python: Tavily API client for web research
• ...

Install

We use make and uv. This command install uv if not present, then load the project dependencies.

• make install

The project requires Python 12 (that can be easily installed by uv)

Configuration:

• Application settings are in file : app_conf.yaml ; Should likely be edited (and improved...)
• API keys are taken from a .env file, in the project directory or its parents

Quick test:

• Run echo "computers" | uv run src/main_cli.py run joke -m fake_parrot_local
  • It should display 'Tell me a joke on computers'
  • Don't care about the warnings
  • add --help to see the different options
  • You can change the LLM by taking one in defined in models_providers.yaml (if keys are in the .env file)
• Run 'make test'
  • There some issues with several tests in //.
  • You might need to change section 'pytest' app_conf.yaml too

Key Files and Directories

Configuration

• app_conf.yaml: Main configuration file for LLMs, embeddings, vector stores and chains.
• models_providers.yaml: Contains model definitions and provider configurations
• pyproject.toml: Poetry project configuration

Core AI Components facilitating LangChain programming

• src/ai_core/: Core AI infrastructure
  • llm.py: LLM factory and configuration
  • embeddings.py: Embeddings factory and management
  • vector_store.py: Vector store factory and operations
  • chain_registry.py: Runnable component registry
  • cache.py: LLM caching implementation
  • vision.py: Facilitate use of multimodal LLM
  • prompts.py: Prompt templates and utilities
  • structured_output.py: Structured output generation helper
  • instrumentation.py: Telemetry and monitoring

Reusable AI Components

• src/ai_extra/: Generic and reusable AI components, integrated with LangChain
  • gpt_researcher_chain.py: LCEL encapsulation of GPT Researcher
  • smolagents_chain.py: SmolAgents implementation
  • mcp_tools.py: Model Context Protocol utilities

Demos with UI

• src/webapp/: Streamlit web application and pages
  • pages/: Streamlit page implementations
  • 1_▫️_Runnable Playground.py: Page to test registered LangChain runnables
  • 2_▫️_MaintenanceAgent.py: a ReAct agent to help maintenance planning
  • 3_▫️_Stock_Price.py: a tool calling agent to ger and compare stock prices
  • 4_▫️_DataFrame.py: a tool calling agent query tabular data
  • 5_▫️_Mon_Master.py: Example of similarity search project
  • 7_▫️_GPT_Researcher.py: Page demonstrating GPT Researcher
  • 9_▫️_SmallAgents : Page to execute SmollAgents
  • 12_▫️_Crew_AI.py: CrewAI demonstration
  • ...

Utilities

• src/utils/: Utility functions and helpers
  • config_mngr.py: Configuration management
  • pydantic/: Pydantic model extensions
  • streamlit/: Streamlit-specific utilities
  • singleton.py: Singleton pattern implementation
  • units.py: Unit conversion utilities

Notebooks for GenAI training

• python/Notebooks

Reusable UI Components

• src/webapp/ui_components/: Generic and reusable User Interface components for Streamlit
  • llm_config.py: Form to select LLM and common configuration parameters
  • capturing_callback_handler.py: Callback handler for Streamlit
  • clear_result.py: State management utilities

Examples (more for training purpose)

• src/ai_chains/: Example chains and RAG implementations. Examples:
  • B_1_naive_rag_example.py: Basic RAG implementation
  • B_2_self_query.py: Self-querying retriever example
  • C_1_tools_example.py: Tool usage demonstration
  • C_2_advanced_rag_langgraph.py: Advanced RAG with LangGraph
  • C_3_essay_writer_agent.py: Essay writing agent
  • ...

Testing and Development

• tests/: Unit and integration tests
• src/demos/: Various demonstration implementations
  • maintenance_agent/: Maintenance planning demo
  • mon_master_search/: Search demo
  • todo/: Task management demos
• Makefile: Common development tasks
• Dockerfile: optimized dockerfile
• CONVENTION.md: Coding convention used by Aider-chat (a coding assistant)