Cold-chain dispatch agent with spatial data and compliance RAG

The scenario

FrostLine turns a legacy cold-chain logistics database into an autonomous dispatch assistant. A ReAct agent queries truck telemetry with spatial SQL, pulls forecasts for the route ahead, and checks the temperature-handling rulebook through RAG to decide whether a load is at risk and what to do about it — reroute, re-ice or flag the receiver. Every query and recommendation lands in an append-only audit table, and the whole system deploys to a VM through a CI/CD pipeline.

System architecture

High-level flow of FrostLine, from input to output.

What you will learn

• Giving an agent safe, read-only spatial SQL tools
• Combining live APIs with document retrieval in one reasoning loop
• Designing an immutable audit trail for agent actions
• Operational alerting for at-risk shipments

What you will build

• Telemetry database with routes, readings and geofences
• Agent with SQL, weather and rulebook tools
• Dispatch console showing risk, reasoning and suggested action
• Audit table and a CI/CD deploy to EC2

Build plan

1. Seed the telemetry database with simulated fleets
2. Build SQL and weather tools with guardrails
3. Index the compliance rulebook
4. Assemble the agent and the console
5. Add the audit trail and deploy

Tech stack