AI risk monitoring

How MortgageAgent watches every funded loan, 24 hours a day, 4 eyes deep

Every position in your portfolio is read by an AI agent the moment it funds and rescored on a published cadence. Below is the architecture in plain language — the signals it reads, the rules that trigger a rebalance, the window it watches from, and the ledger that keeps the audit trail.

Try the scorer
The risk architecture

Intake signals, rebalance rules, the monitoring window, and the ledger that holds it

Four parts, one continuous loop — the same model that scored a loan at intake is the one watching it today, and the same ledger it writes to is the one your auditors will read.

01
Intake signals the agent reads on every funded loan
On funding, the agent ingests the full origination packet — borrower identity, property valuation, LTV/CLTV, DSCR, lien position, title, and the originator’s underwriting packet. Borrower financials, credit pulls, and the appraisal report are stored alongside the loan record so the model can re-score without re-pulling. Sourced loans are already originator-vetted, but every funded position is rescored against the published risk model before it lands in a portfolio.
02
Rebalance rules & cadence
Every funded position is rescored against the live risk model on a scheduled cadence — fresh signals on application, at wire confirmation, and again on each scheduled run declared in polsia.toml. When a position breaches a threshold (LTV drift, valuation pullback, delinquency flag, lien change) the agent proposes a rebalance: trim exposure, exit the position, or hold with elevated monitoring. Originators see the same recommendation the investor dashboard does — the model is one source of truth.
03
The 24/4 monitoring window
Active positions are watched around the clock by scheduled background jobs declared in polsia.toml. Intraday jobs re-score drift, value swings, and servicer reports; nightly jobs reconcile the ledger against the custodian and originator feeds. Alerts surface on the dashboard the moment a position crosses a threshold, and the same record is available to investors on every portfolio view — the public scorer on a sample loan is the same engine watching your positions.
04
Custody of the audit trail — Postgres
Every signal, score, rule trigger, rebalance decision, and custodian reconciliation is written to a Postgres-backed loan ledger — the same record the loans dashboard reads. Each row carries a monotonic ID and a timestamp, so an outside auditor can replay any position’s history from funding through exit. The ledger is the single source of truth for compliance, investor reporting, and the operator-facing admin views; no score, alert, or recommendation lives outside it.

Want the broader picture? Read the three-step portfolio loop or run the partner program for how originations feed the pipeline.

Talk to the desk

See the agent run on a real portfolio

Every funded loan starts the same way — accredited capital routed into a position the model clears, then watched 24/4. Apply through the existing accreditation intake, partner with the originator network feeding the ledger, or write to the risk desk directly.