B3
Track B — Multi-Agent Orchestration & Decision Support

Campus/Facility Infrastructure Decision-Support Agent

Systems where multiple specialized agents collaborate, hand off tasks, and reach a combined decision.

Background

When equipment (AC units, generators, elevators) breaks down, facility managers often rely on memory instead of reviewing what happened with similar equipment before — wasting time and missing recurring patterns.

Objective

Develop an AI agent that helps facility managers diagnose new issues by retrieving similar past maintenance cases and recommending a likely cause and fix, with clear reasoning.

Key Features

1. Searchable knowledge base of past maintenance records 2. Similar-case retrieval for new complaints 3. Diagnosis reasoning based on retrieved cases 4. Fix recommendation with estimated cost/time and urgency 5. Plain-language explanation of the reasoning chain

Expected Solution

A semantic search layer over historical maintenance logs A diagnosis agent that reasons over retrieved similar cases A recommendation agent for fix steps and urgency A simple logging/dashboard interface

Suggested Technology Stack

Retrieval: FAISS / ChromaDB (semantic similarity search) Reasoning: LLM + LangGraph (retrieval → diagnosis → recommendation pipeline) Data: Synthetic dataset of 200–300 maintenance records Frontend: Simple web dashboard

Expected Outcomes

Faster, more consistent diagnosis of recurring equipment issues Reduced downtime through pattern-based recommendations An explainable decision-support pattern for maintenance teams

Possible Use Cases

A campus facilities team triaging a new AC failure A hospital diagnosing repeated equipment issues A factory logging and prioritizing maintenance requests

Evaluation Parameters

Relevance of retrieved similar cases Soundness of diagnosis reasoning Usability of the dashboard Handling of multiple simultaneous complaints (if implemented)