From Reactive to Proactive Mobility Insights: Caltrans’ Agentic Solution on a Unified Lakehouse
Overview
| Experience | In Person |
|---|---|
| Track | Artificial Intelligence & Agents |
| Industry | Enterprise Technology, Public Sector, Transportation |
| Technologies | Delta Sharing, Unity Catalog, Lakebase |
| Skill Level | Beginner |
| DOWNLOAD SESSION SLIDES | |
Caltrans is transforming Traffic Mobility Insights (TMI) into an agent-forward transportation AI platform built on the Databricks lakehouse and lakebase. At its core is an agentic architecture that unifies structured telemetry, geospatial data, and unstructured documents within a governed Delta foundation—eliminating fragmented pipelines and legacy database silos. Lakebase enables high-performance operational lookups and spatial workloads, while the lakehouse provides scalable analytics, model training, and data sharing. A centralized AI switchboard—powered by Databricks Model Serving and MLOps—routes intent across domain agents, enforces policy guardrails, orchestrates tools, and manages state and evidence. This is decision intelligence, not a chatbot. The future state evolves to a federated, Agent-as-a-Product model, where interoperable agents collaborate under a centralized AI control plane—delivering secure, explainable, and proactive transportation management at statewide scale.
Session Speakers
Ajali Sen
/Director
Accenture
Duper Tong
/Chief, Division of Traffic Operations
Caltrans
Full Summary
Inside Caltrans: How California Built an Agentic Mobility Intelligence Platform on Databricks
California's Department of Transportation unified sprawling transportation data and turned it into a production-ready, agent-driven decision platform called Traffic Mobility Insight (TMI). Built on the Databricks Platform with Lakehouse and Lakebase, the system grounds multimodal AI experiences in the same governed data foundation that powers analytics and operations. The approach reflects a broader pattern explored in how companies winning at AI built their data layer first, where a disciplined foundation rather than surface-level prompting drives lasting results.
FAQ
TMI is a multi-agent platform that lets engineers ask natural language questions and receive maps, dashboards, reports, and countermeasure recommendations. It supports safety analytics, freight routing and permitting, emergency evacuation planning, and faster access to engineering guidance across a unified, governed data foundation.
Scale and consolidation drove the choice. Standalone PostgreSQL could not reliably handle 66 billion records and added governance, security, and monitoring overhead. Lakebase delivers low-latency reads, native PostGIS support, and seamless sync from gold-layer data under the same Unity Catalog governance as analytical workloads.
Unity Catalog enforces governance at the data asset level, embedding classification, ownership, retention, and sharing into metadata. Agents inherit those controls automatically, which enabled the solution to expand beyond the initial four use cases while maintaining consistent security and lineage.
No. A spatial analytics agent converts plain-English prompts into spatial SQL, and ArcGIS renders interactive maps. Results align to Caltrans post-miles, so insights drop directly into existing traffic management workflows without extra translation.
A statewide congestion report that took about 1,000 staff hours now generates in roughly two seconds. Statewide incidents surface within 20 to 25 minutes, and emergency routes account for closures. Platform metrics include a 17 percent faster query response, a 52 percent reduction in compute costs, and a 50 percent improvement in time to insight.