JourneyMesh: A Multi-Agent LLMOps Platform | IIT Guwahati Term Project

JourneyMesh is my Term Project for the BSc (Honours) Data Science and Artificial Intelligence program at IIT Guwahati. JourneyMesh is a multi-agent LLMOps platform for AI-assisted travel planning. It uses a supervisor-driven workflow to dynamically select specialist agents for flights, hotels, weather, budgeting, transportation, and itinerary planning. The project demonstrates how agentic AI can be combined with state management, external tools, guardrails, evaluation, human-in-the-loop review, observability, persistence, and production deployment. Key technologies used: • LangGraph and LangChain for multi-agent orchestration • Model Context Protocol (MCP) for local and remote tool integration • FastAPI and Pydantic for the backend API and validation • PostgreSQL for application data and workflow checkpoints • React for the frontend • LangSmith for optional tracing and observability • Docker and Docker Compose for containerization • GitHub Actions and GHCR for CI/CD • OVHcloud VPS and Caddy for deployment The system also includes dynamic agent routing, dependency-aware revisions, tool guardrails, human approval, deterministic evaluation, MCP failure handling, and offline provider support. The evaluation suite currently contains nine test scenarios covering normal travel requests, selective agent routing, multilingual responses, revisions, invalid inputs, prompt injection, and other adversarial cases. This project helped me explore not only LLM prompting, but also the engineering challenges behind production-style agentic systems, including workflow control, tool security, persistence, evaluation, failure recovery, and deployment. Project: JourneyMesh Program: BSc (Honours) Data Science and Artificial Intelligence Institution: Indian Institute of Technology Guwahati (IIT Guwahati) Term Project - Trimester 9 #JourneyMesh #LLMOps #AgenticAI #MultiAgent #LangGraph #MCP #FastAPI #PostgreSQL #GenerativeAI #IITGuwahati