Skip to content
TechFabric
AutoFabric

Automotive finance · Our own product

The auto loan lifecycle, run by governed agents

Origination and refinance steps that eat the calendar, run by agents on Databricks: credit pulls, VIN decode, deal structuring, stipulations and payoff quotes, with a policy engine deciding what each agent may do and every decision keeping its lineage.

Client
AutoFabric
Sector
Automotive finance · Our own product
Stack
Unity CatalogMosaic AIDatabricks AppsTemporal

Challenge

The steps that eat the calendar are the ones nobody wants to own

Origination and refinance are long processes with a regulator waiting at the end of them. Credit pulls against Experian and TransUnion, VIN decode, deal structuring, chasing stipulations, payoff quotes: each one is routine, each one is slow, and together they set how long a deal takes.

Automating them is the obvious move and the dangerous one. A lender who lets software make those calls has to be able to show, months later, why a given deal went the way it did, and to prove that TILA, ECOA and FCRA were satisfied on every action rather than sampled once a quarter.

Strategy

Agents that do the work, policy that decides what they may do

Separate the two questions. An agent runs the step; a policy engine decides what that agent is allowed to touch and under whose permissions. Compliance is evaluated on every action as part of the run, so the audit trail is a by-product of doing the work.

That only holds if the long steps survive real conditions. A payoff quote that takes three days cannot restart from the beginning because a process was recycled overnight.

Solution

The loan lifecycle, run by governed agents on Databricks

Origination and refinance agents run the credit pulls, VIN decode, deal structuring, stipulation chasing and payoff quotes. An OPA policy engine gates every action, with TILA, ECOA and FCRA checks built in as policy the agent is measured against.

Unity Catalog holds the governed data and the permissions the agents inherit. Mosaic AI serves the models. The operator surfaces are Databricks Apps, running in the workspace. Temporal carries the long-running steps, so a quote that takes days survives a restart.

Integration follows the deals rather than the architecture diagram: DealerTrack, RouteOne and CDK, because that is where the deals already are.

Results

A decision you can defend nine months later

Every decision keeps its lineage. When an examiner asks why a deal went the way it did, the answer is a record rather than a reconstruction.

AutoFabric is a product we sell and operate, which is why the architecture here is one we can open rather than one we recommend from a slide.