GMF Forge AI Monorepo

Enterprise-grade AI platform for building RAG applications and AI agents, organized as independently deployable packages.

🏗️ Architecture

This monorepo contains six independently versioned and deployable packages:

gmf-forge-ai-shared-core       (v1.0.0) - Foundation: LLM gateway, registries, tools
gmf-forge-ai-shared-enterprise (v1.0.0) - Enterprise: Observability, governance, evaluation
gmf-forge-ai-ux               (v1.0.0) - User interfaces and components
gmf-forge-ai-data             (v1.0.0) - RAG, retrieval, embeddings, connectors
gmf-forge-ai-orchestration    (v1.0.0) - Agents, workflows, multi-agent systems
gmf-forge-ai-infra            (v1.0.0) - Infrastructure and deployment

🎯 Use Cases

RAG Application (Basic)

pip install gmf-forge-ai-shared-core gmf-forge-ai-data gmf-forge-ai-ux

AI Agent Application (Basic)

pip install gmf-forge-ai-shared-core gmf-forge-ai-orchestration gmf-forge-ai-data gmf-forge-ai-ux

Enterprise Applications

Add gmf-forge-ai-shared-enterprise for: evaluation, governance, PII detection, cost tracking, compliance logging.

🚀 Quick Start

For Users (Installing Packages)

# Install only what you need
pip install gmf-forge-ai-shared-core gmf-forge-ai-data gmf-forge-ai-ux

# Use in your application
from gmf_forge_ai_shared_core.llm_gateway import UnifiedLLMGateway
from gmf_forge_ai_data.retrieval import HybridRetriever
from gmf_forge_ai_ux.chat import ConversationUI

🏗️ Architecture & Dependencies

The monorepo follows a clean, acyclic dependency architecture where higher layers depend on lower layers, but never the reverse.

Dependency Graph

Layer Responsibilities

🔹 shared-core (Foundation - No dependencies)

• LLM Gateway (unified Azure OpenAI access)
• Observability (logging, tracing, metrics)
• Configuration management
• SSL certificate handling
• Base abstractions and utilities

🔹 shared-enterprise (Optional - Depends on: shared-core)

• Model evaluation frameworks
• Governance and compliance
• PII detection and redaction
• Cost tracking and analytics
• Audit logging

🔹 data-layer (RAG Pipeline - Depends on: shared-core)

• Embeddings (Azure OpenAI, batching, caching)
• Chunking (7 strategies: fixed, semantic, recursive, sentence, markdown, code, combined)
• Vector stores (Azure AI Search, in-memory) + Retrieval (8 strategies)
• Query optimization
• Context processing
• Indexing pipelines
• Data connectors
• Memory management

🔹 orchestration-layer (Agents - Depends on: shared-core, data-layer)

• Agent frameworks
• Tool integration
• Multi-agent coordination
• Workflow orchestration
• Task planning and execution

🔹 ux-layer (Interfaces - Depends on: shared-core)

• Chat interfaces
• Streamlit components
• Gradio integrations
• UI utilities

🔹 infra-layer (Infrastructure - Depends on: shared-core)

• Terraform/Bicep templates
• Deployment scripts
• Configuration templates
• Infrastructure monitoring

🎓 Example Applications

See the apps/ directory for complete example applications:

• basic-rag-app/ - Simple RAG application
• react-agent-app/ - ReAct agent implementation
• multi-agent-system/ - Multi-agent orchestration
• policyhub-agent/ - Policy hub agent with RAG-backed Q&A
• task-agent-template/ - Reusable template for building task agents

🔌 MCP Servers

Model Context Protocol servers for tool integrations:

• search-server/ - Web and document search capabilities
• database-server/ - SQL database query execution
• filesystem-server/ - File system operations
• api-server/ - Generic REST API interactions
• policyhub-mcp-server/ - Policy hub domain tools and retrieval

See mcp-servers/README.md for setup and usage.

🤝 Contributing

See CONTRIBUTING.md for detailed guidelines.