project · 2025-2026
Traffic-analytics agent
Designed a multi-agent prototype for answering traffic-analysis questions with scoped tools, structured evidence, and a single grounded response.
The problem
Traffic-analysis questions can require routing, historical comparison, incident context, and regional analysis. One general-purpose prompt is difficult to make both reliable and explainable.
Approach
- Split the workflow into focused agent roles with scoped tool surfaces.
- Used structured intermediate outputs so the coordinator can compose an evidence-backed response.
- Applied validation and guardrails at tool boundaries to keep actions within their intended scope.
How it works
This scripted example uses fictional observations to show why specialised roles can help: each agent handles one evidence-gathering task, and a coordinator combines the results once.
Outcome
The prototype illustrates when a multi-agent design is useful: each role has a distinct responsibility, while the user receives one clear, grounded answer.