About
We are building mortgage intelligence, and starting where the checking already happens.
riTara.ai is building specialised intelligence for mortgage documents, policies, calculations and exceptions. E3 is where that work meets real files today: quality control, run on whole loan files, with the evidence kept behind every finding.
The problem we are actually working on
Every decision in this industry is a judgement about a stack of documents
Whether to lend. Whether to buy the loan. What to service, and on what terms. What sits inside a security. Whether a file survives an examination in three years. Each of those is a question about what a few hundred pages actually say, and each is answered today by a person reading a fraction of them against a policy that keeps moving.
That is the layer the industry has never had: something that reads mortgage documents, understands what they assert, knows the policy that applies, and can prove the answer to whoever asks later. It serves the originator, the aggregator, the investor, the servicer and the auditor, because they are all asking questions of the same documents.
We start with quality control because it is a point where the checking already happens, the answer is already supposed to be evidenced, and the work is measurable. It is the first application, not the ambition.
Why quality control first
The first review by someone with no interest in the deal closing
A defect in a loan file is usually described from one end of the chain: a repurchase demand, an audit finding, a cost. That framing is accurate and narrow.
The same defect at the other end is a borrower whose income was assessed incorrectly, a disclosure that arrived late, or a loan that should not have closed on those terms. The borrower has the most at stake in the file being right and the least ability to check any of it.
Post-closing quality control is one of the few reviews performed by someone with no direct interest in the transaction having closed. Making that examination complete rather than sampled, and reproducible rather than remembered, improves the accuracy of lending for every party at once.
Where we are going
A mortgage-domain model, and what it has to be better at
Today E3 runs a general-purpose vision-language model behind a versioned rule engine, with a person resolving the exception. That is where we are today.
We are building a mortgage-domain model for the documents, the policy and the exceptions this industry actually produces. Not to read pages better, which is a capability the whole market is buying from the same few providers, but to reason about what it has read: which qualifying-income method a programme requires, when a difference between two documents is legitimate, which document should exist and does not, and how to state an unresolved exception so a reviewer can act on it.
The rule engine is what earns the right to that model, because it produces the signal that is otherwise unavailable: which rule fired, what a reviewer did with it, and why. That signal improves the rules first an override means a rule was wrong, too broad or badly explained, and the answer is a reviewed new rule version rather than a silent model update.
Customer documents are not training data. A customer's policy configuration and overrides stay in their tenant, and nothing derived from them becomes reusable beyond it without expert validation and the customer's written agreement.
"Industry-leading" becomes publishable only alongside a stated benchmark, a stated document set and a comparison anyone can reproduce. We will publish the method with the number, or we will not publish the number.
Leadership
Shailesh Bhujbal, Co-Founder
Nearly three decades in fintech, building document and data platforms at scale across mortgage finance, cloud infrastructure, international development and industrial engineering. The problem has been the same throughout: turning documents into systems whose output can be relied on, in environments where the record has to survive examination years later. E3 is the continuation of that.
Talk to us about your files.
A working session on a synthetic file, or de-identified files of your own. No obligation and no procurement process required.