Skip to content

arnavanuj/mcp-ticket-rag

Repository files navigation

MCP Ticket RAG

An AI-powered system for querying and summarizing GitHub issues and comments using a hybrid MCP live retrieval + RAG architecture, enabling structured issue lookup and semantic reasoning over ticket data.


Technology Stack

  • Orchestration: Python service layer with modular routing logic
  • UI Layer: Streamlit interactive interface
  • LLM Runtime / Models: Ollama
    • phi3-mini – semantic resolver
    • mistral – answer generation
    • llava-phi3 – vision / OCR processing
  • Embeddings: Sentence Transformers (all-MiniLM-L6-v2)
  • Vector Store: ChromaDB
  • Data Access: GitHub MCP server (live issue and comment retrieval)
  • Language: Python

AI & System Design Patterns

  • Retrieval Augmented Generation (RAG) for semantic issue search
  • Hybrid MCP + RAG architecture combining live API retrieval with vector search
  • Semantic query routing using a lightweight LLM resolver
  • Modular AI service design with independent components
  • LLM orchestration pipeline separating reasoning and generation models
  • Streaming LLM execution with observability diagnostics
  • Tool-assisted retrieval via MCP server integration

Capabilities Demonstrated

  • Building LLM-driven developer productivity tools
  • Designing hybrid retrieval architectures (live APIs + vector search)
  • Implementing multi-model orchestration pipelines
  • Integrating LLMs with external systems using MCP
  • Building observable AI services with request diagnostics
  • Developing modular AI engineering systems suitable for production extension

High-Level Flow

User → Streamlit UI → Semantic Resolver (phi3-mini) → Router →
MCP Live Retrieval / Vector RAG → Answer Generation (Mistral) → Response


Future Plans

  • Introduce multi-agent workflows for issue investigation and root cause analysis
  • Add automated issue triage and classification agents
  • Improve context compression and retrieval ranking for faster RAG responses

About

Hybrid MCP + RAG system for querying and summarizing GitHub issues using local LLMs, vector search, and semantic routing via Ollama. Capability Highlight: Demonstrates hybrid AI retrieval architecture combining MCP live APIs, vector RAG, and multi-model orchestration (phi3-mini + Mistral) for developer productivity tools.

Topics

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors

Languages