Finalist
Autonomous Infrastructure Auditor
1 in 4 Americans lives with a disability, yet most cities have no systematic way to audit pedestrian infrastructure for ADA compliance at scale.
Core trick
Specialist agents compute slope from terrain elevation data and pair it with Street View vision checks against ADA code sections.
Description
1 in 4 Americans lives with a disability, yet most cities have no systematic way to audit pedestrian infrastructure for ADA compliance at scale. Manual inspections are slow, expensive, and cover only a fraction of the street network.
The Autonomous Infrastructure Auditor solves this by deploying a fleet of AI specialist agents across any urban district to autonomously detect accessibility violations such as steep ramps, missing tactile indicators, insufficient crosswalk timing. These are things that prevent wheelchair users, visually impaired pedestrians, and ambulatory individuals from navigating the city safely.
A District Orchestrator dispatches three specialist agent types concurrently, each walking real geodetic terrain and applying persona-specific ADA thresholds to measured physics.
What makes the system distinctive is how it grounds every finding in multiple modalities simultaneously: real terrain elevation data drives the slope physics, live Street View imagery is analyzed by Gemini's vision capability to visually confirm hazards at the exact coordinate, and the agents' regulatory reasoning is anchored to specific ADA code sections rather than generalized descriptions.
When a location fails multiple criteria at once, two specialist agents negotiate a unified fix through bidirectional context passing, producing a single remediation that resolves compound hazards together.
The result gives city planners actionable, location-specific violations at a scale and speed no human audit team could match.
README excerpt
# Autonomous Infrastructure Auditor ### San Francisco Digital Twin — Nob Hill ADA Compliance Fleet A multi-agent system that deploys 1,000 AI specialist agents across Nob Hill, San Francisco to autonomously detect ADA accessibility violations. Agents walk real geodetic terrain, negotiate with each other over compound hazards, self-correct their regulatory findings, and ground every infraction in real Google Street View imagery and AI-generated engineering blueprints. --- ## What It Does The system simulates a district-wide accessibility audit at a scale no human team could execute. A District Orchestrator Agent dispatches a fleet of 1,000 specialist sub-agents concurrently across a 600-meter radius of Nob Hill. Each agent walks a 50-step path through real San Francisco terrain, applies ADA compliance thresholds specific to its disability persona, and generates regulatory findings backed by actual ADA code sections. The results are visualised on a live 3D Google Maps WebGL dashboard with colored agent markers, interactive infraction modals, real Street View photos of each flagged location, and AI-generated CAD blueprints for each infraction type. --- ## Agentic Architecture ### District Orchestrator Agent The orchestrator is the only stateful object in the system. It owns the territory definition, manages an LLM response cache, and dispatches the entire fleet via `asyncio.gather` — all 1,000 sub-agents start concurrently. A `Semaphore(10)` rate-limits concurrent Gemini