ValonInterview Hub
The product · ValonOS

One unified system,
compounding leverage.

ValonOS is the AI-native operating system for regulated finance. Workflows, data, money movement, and compliance — running on the same source of truth. We are long AI progress. We are not long the AI competition.

What ValonOS does

Five primitives. One source of truth.

Every product surface, every workflow, every AI agent — built on the same five primitives. No shadow systems, no reconciliation hell.

Bitemporal ledger

Captures regulatory change correctly. Events are immutable but reversible.

Composable workflows

Servicing actions as APIs. Compose into workflows. Hand any unit to a human or an agent.

LLM-ready actions

Every action is designed to be performed by either humans or agents, with the audit trail to prove it.

Compliance as code

50 states, GSEs, CFPB, SOC, USAP — encoded into the platform, not bolted on.

Money movement

Cash and accounting unified with operations. The single ledger no incumbent has.

Two product surfaces

Built for the people who touch the loan.

All of the platform lives behind APIs. Two consoles ride on top — one for homeowners, one for servicing operators. Both share the same source of truth.

Servicer console
Servicer console

The control room for operations.

Workflows, task queues, audit trails. Every action is captured, replayable, and ready for an agent to take over.

Homeowner console
Homeowner console

The mortgage you actually want to log into.

Pay your bill, manage escrow, get the answer in one click instead of three calls. 90% CSAT, by design.

AI in production · Not a demo

An example: matching the unidentifiable check.

Every day, mortgage servicers receive checks with no clear loan reference. Traditionally, an ops person spends 10–20 minutes hunting through systems. On ValonOS, the agent reads the check, proposes a match with confidence, and waits for human approval.

Unidentified check scan
Step 1

Vision model parses the check.

Routing number, account, memo line, signature — structured into our action graph.

AI suggested match
Step 2

The agent proposes a match.

Confidence scored against borrower history, escrow signals, and ledger state. Operator approves in one click.

Every servicing action on ValonOS is built so a human or an agent can perform it — with the same audit trail.

That isn’t a feature flag. It’s a foundational design decision that took six years of modeling mortgage operations from first principles. It’s also why a workflow that takes 20 minutes today can take 30 seconds next quarter.

We use this internally too. Nicolebot is the cross-functional agent that runs in our Slack — querying the data graph in natural language to debug loan-level issues, tagging engineering threads, opening pull requests, and beating on-call engineers to the first response on incidents. Used by nearly the whole company.

The stack

Boring tech. Hard problems.

The stack is deliberately unsexy. The complexity that deters casual innovators is music to us — conquering it is the moat. Our differentiation isn’t what we wrote it in. It’s the ledger that handles regulatory change played forward with correct accounting, the loan-state model that represents every product variation across 50 states, and the six years it took to build them.

Surfaces
Homeowner consoleServicer consolePartner APIsMarketplace
AI & automation
LLM agentsWorkflow engineAction graphAudit log
Core servicing APIs
PaymentsEscrowInvestor reportingLoss mitigationCompliance
Source of truth
Bitemporal ledgerEvent timelineMoney movement
Frontend
React · TypeScript
API layer
GraphQL
Backend
Python
Database
MySQL · bitemporal
Infrastructure
Google Cloud Platform
Models
Anthropic · OpenAI · in-house
Engineering at Valon
“Huge industries still run on software from the 1980s. Financial services, insurance, medicine — software abandoned them, so they built huge operations teams to fill the gap. This stuff should be software. We’re the only company creating the system of record by deeply understanding the industry and building trust — the two gaps no one else closes.”
Jake Mintz · Chief Product Officer