AC
Founder
Arya Chudasama
For
leveraged-finance credit teams at banks & NBFCs
The credit analyst that never leaves the call.
It carries context through every stage of a deal: pre-call, live on the call, and after, applying your credit reasoning in real time instead of transcribing after the fact.
MR
M. Reyes · CFO, Atlas
DA
D. Arora · Sponsor
M. Reyes
100+
transactions analyzed
35+
credit rules encoded
3
stages of context
0
added call latency
PRE-CALL
Loads the context
Ingests the CIM (Confidential Information Memorandum), prior filings and your open credit questions so it walks in already knowing the story and what still needs answering.
LIVE
CALL
Reasons in real time
Tests each claim against your rules as it is said, surfaces the follow-up while management is still on the line, and flags what stays unquantified.
POST-CALL
Writes the record
Produces a defensible, sourced summary mapped to your IC (Investment Committee) template, every flag and implication carried straight through.
The old way
Notes get written up after the fact. Context is reconstructed from memory, the sharp follow-up never gets asked, and the record is only as good as who was in the room.
The new way
Context is carried live. Every claim is reasoned against your rules as it's said, the follow-up is surfaced on the call, and a defensible record writes itself as you go.
The workflow
From the data room to the IC memo, in one call.
A borrower brings a facility request to the desk. The analyst runs the call on Underwriter AI, and every answer is questioned, checked and sourced on the way to committee.
ML
Borrower
Meridian Logistics
FacilityTerm loan
SectorLogistics
Ask₹120 Cr
Credit analyst
Bank desk
Live analysis
Underwriter AI
Underwriter AI
Questions drawn from the data room
Prompts surface live, on the call
Counter-questions on weak answers
Answers checked against the filings
IC-ready memo for the desk
IC memo delivered
Applies credit reasoning
It thinks in covenants, leverage and cash conversion, not words per minute. Transcription is table stakes; judgment is the point.
Catches what gets missed
The unasked follow-up, the number that was dodged, the claim that quietly contradicts the last call, surfaced, not lost.
Builds a defensible record
Every conclusion is sourced back to what was actually said, so the memo holds up in committee and in review.
Trust
Early product, built for institutional use.
It's decision-support with a human in the loop, never an autonomous decision-maker. Here's exactly what it does today and what's next.
TODAY
✓Live reasoning against your own encoded credit rules
✓Sourced, IC-ready summaries
after every call
✓Flags for unquantified guidance and contradictions
✓Human-in-the-loop on every output
NEXT
›Deeper integration with
your document stack
›Portfolio-level pattern
detection across calls
›Configurable rule
authoring for your team
›Audit trail export for
compliance review
Security & compliance
Built to sit inside your perimeter.
Credit data is sensitive, and we treat it that way. Here is how deployment, data handling and the regulatory posture work, so your risk and compliance teams have their answers up front.
On-premise & in-your-perimeter deployment
Underwriter AI is built to run inside your own environment, not on a shared multi-tenant service. Model inference uses AWS Bedrock, where inputs are not retained and not used to train models, and a region-pinned mode keeps processing inside a chosen region. No MNPI has to leave your control.
DATA
HANDLING
Your data stays yours
Inference runs on Bedrock with no retention and no training on your inputs. Region-pinning is available for data-residency requirements, and product demos use only synthetic data.
AUDIT
TRAIL
Tamper-evident by design
Every question, answer, flag and confidence score is hash-chained (SHA-256) to the record before it, so altering any row breaks the chain. It is shaped for SEBI's structured-digital-database expectations.
CONTROL
Human-in-the-loop, always
A deterministic rule layer runs before the model, and low-confidence findings are surfaced for analyst judgment rather than asserted. Nothing the product outputs approves, prices or declines a deal.
REGULATORY
POSTURE
Honest about the gaps
We deploy as material technology outsourcing under IFSCA / RBI, and treat DPDP Act duties as pre-deployment work. Enterprise controls such as access control, tenant isolation and redaction are in progress; we would run a supervised sandbox pilot before production.
Team
The people building
Underwriter AI
JD
Engineer
Engineer
Jay Dobariya
HR
Researcher
Researcher
Harsh Rao
VS
Engineer
Vrushti Somaiya
JT
Strategy
Strategy
Jignesh Thakkar
FAQ
Questions credit teams ask first.
No. It is decision-support with a human in the loop. It surfaces reasoning and questions faster; a qualified analyst still owns every credit decision.
Your rules are encoded and applied in real time. It tests each claim against them rather than against a generic model of what a call should contain.
Built for institutional use. Data stays scoped to your workflow with a human-reviewed audit trail. Deeper compliance export is on the near-term roadmap.
No added latency to the call itself. Analysis runs alongside and surfaces to the analyst, not to the room.
Yes. It is built to deploy inside your perimeter rather than on a shared service. Inference runs on AWS Bedrock with no retention and no training on your inputs, and a region-pinned mode keeps data in a chosen region.
No MNPI has to leave your control. We use Bedrock precisely because inputs are not retained or used for training, and today's demos run only on synthetic data.
A deterministic rule layer filters every statement first, with no AI, and narrows each moment to a handful of relevant rules. Only that small subset goes to the model, which must return a structured result. Less AI surface, more control and auditability.
Every finding carries a confidence score, and low-confidence items are shown for analyst judgment rather than asserted. Nothing it outputs approves, prices or declines a deal; a qualified human always decides.
Bring a live deal or a recent recorded call. We run it through and show you exactly where the reasoning holds and where it flags.
Put it in a real workflow.
Bring a live deal or a recent call. We'll run it through and show you where the reasoning holds, and where it flags.
Discuss a pilot
Arya (Founder & CEO)[email protected]
General[email protected]