An intelligent Multi-Agent system that demystifies VAR decisions and provides tactical insights using explainable AI. Built with IBM Granite, Docling, and LangFlow for the IBM Skills Build AI Builders Challenge
This project employs an advanced RAG pipeline that utilizes IBM Docling to process FIFA rule PDFs and store them in a FAISS vector database. It then leverages semantic search to quickly extract the most relevant rules, providing them to the user as cited sources
Key Tech: RAG Pipeline | IBM Docling | FAISS Vector DB
The core intelligence of the system is driven by the IBM Granite language model, which equips the agents with natural language understanding and data analysis capabilities. It transforms complex tactical data and rigid VAR regulations into clear, transparent explanations supported by reasoning.
Key Tech: OIBM Granite | Data Analysis | LLM Reasoning
The system is built on a Multi-Agent architecture featuring an Orchestrator that routes queries in parallel to specialized agents (VAR-Lens and Tactical Pulse). A key strength of this design is the Tool-Agent separation, which ensures the system remains highly modular, flexible, and scalable.
Key Tech: Multi-Agent | Orchestrator | Parallel Routing
The system’s Orchestrator autonomously detects user intent, delegates tasks to the appropriate agents, and synthesizes the final response without any manual intervention. Furthermore, the development process utilized AI tools like IBM Bob for coding automation and streamlined debugging.
Key Tech: Centralized Governance | AI Tools
