Kita builds AI credit assessment software for lenders operating in markets where borrowers do not fit neatly into traditional bureau data. The platform ingests messy application documents like bank statements, e-wallet histories, and utility bills, uses vision language models to extract fraud-checked signals, and produces a cited credit memo that underwriters can sign or override against their own credit policy. Its AI agents also close gaps in an application, flagging inconsistencies and messaging borrowers directly for missing information before a file lands with a human reviewer.
The launch coincides with a $4.5M round led by Box Group with participation from Y Combinator, announced alongside a milestone of more than $130M in loan volume processed for lenders across the US, Southeast Asia, and Latin America. For lending teams that today rely on manual document review to serve thin-file customers, the pitch is a shift from days of back-and-forth to underwriting decisions in under a minute, with each decision feeding back into the models to sharpen future risk and fraud detection.
Kita is a Y Combinator W26 company based in San Francisco, co-founded by Carmel Limcaoco and Rhea Malhotra, who met before Stanford and have been building together ever since . Carmel, the CEO, is from Manila, a repeat founder, and spent three years in product at Apple, while Rhea, the CTO, has a research background in computer vision , giving the team a combination of emerging-market lending context and technical depth aimed squarely at the document layer that has kept billions of would-be borrowers invisible to formal credit.
Comments (9)
Sign in to join the discussion.
N
Nadia BekovaAug 19, 2026
Sub 60 second underwriting is cute until the compliance team wants an audit trail for every decision. Do lenders get a reason code or just vibes?
T
Tolu AdekunleAug 19, 2026
$130M processed but I want to know the default rate on loans your model greenlit versus the incumbent scorecard. That's the only number that matters here.
M
Marco P.Aug 19, 2026
The launch video pacing was actually clean, no fake keyboard typing shots, no dramatic zoom on a laptop. Rare restraint for a YC reveal.
Y
Yves LambertAug 19, 2026
That screenshot in the tweet has three different font weights fighting each other. The product might be fast but the type hierarchy is not.
P
Priyanka R.Aug 19, 2026
Underwriting in Lagos and Lima with the same model? I need to see the feature engineering because those risk profiles do not rhyme.
S
Samir QureshiAug 19, 2026
Before my risk team touches this: SOC2 Type II, model governance docs, and a data residency story for the EU. Otherwise it dies in procurement.
K
Kenji ItoAug 19, 2026
Every AI credit demo looks incredible on the curated borrower. Show me the one where the model was confidently wrong and what happened next.
A
Ana CzajkaAug 19, 2026
Naive question from someone who has never underwritten anything: what happens in minute 61 when the borrower's bank statement PDF is a photo of a photo?
B
Bea OnwukaAug 19, 2026
Remind me why we moved on from FICO plus a spreadsheet? Worked fine in 2019 and nobody had to explain a neural net to a regulator.
Work With Us
Book an intro call to explore our suite of services. We might have a waitlist.
Kita
Underwrite borrowers anywhere in the world,anywhere in the world, in minutes.in minutes.
Underwrite borrowers anywhere in the world,anywhere in the world, in minutes.in minutes.
Tags
TLVC Rating
Community Rating
No ratings yet
to rate this launch.
Kita builds AI credit assessment software for lenders operating in markets where borrowers do not fit neatly into traditional bureau data. The platform ingests messy application documents like bank statements, e-wallet histories, and utility bills, uses vision language models to extract fraud-checked signals, and produces a cited credit memo that underwriters can sign or override against their own credit policy. Its AI agents also close gaps in an application, flagging inconsistencies and messaging borrowers directly for missing information before a file lands with a human reviewer. The launch coincides with a $4.5M round led by Box Group with participation from Y Combinator, announced alongside a milestone of more than $130M in loan volume processed for lenders across the US, Southeast Asia, and Latin America. For lending teams that today rely on manual document review to serve thin-file customers, the pitch is a shift from days of back-and-forth to underwriting decisions in under a minute, with each decision feeding back into the models to sharpen future risk and fraud detection. Kita is a Y Combinator W26 company based in San Francisco, co-founded by Carmel Limcaoco and Rhea Malhotra, who met before Stanford and have been building together ever since . Carmel, the CEO, is from Manila, a repeat founder, and spent three years in product at Apple, while Rhea, the CTO, has a research background in computer vision , giving the team a combination of emerging-market lending context and technical depth aimed squarely at the document layer that has kept billions of would-be borrowers invisible to formal credit.
Comments (9)
Sub 60 second underwriting is cute until the compliance team wants an audit trail for every decision. Do lenders get a reason code or just vibes?
$130M processed but I want to know the default rate on loans your model greenlit versus the incumbent scorecard. That's the only number that matters here.
The launch video pacing was actually clean, no fake keyboard typing shots, no dramatic zoom on a laptop. Rare restraint for a YC reveal.
That screenshot in the tweet has three different font weights fighting each other. The product might be fast but the type hierarchy is not.
Underwriting in Lagos and Lima with the same model? I need to see the feature engineering because those risk profiles do not rhyme.
Before my risk team touches this: SOC2 Type II, model governance docs, and a data residency story for the EU. Otherwise it dies in procurement.
Every AI credit demo looks incredible on the curated borrower. Show me the one where the model was confidently wrong and what happened next.
Naive question from someone who has never underwritten anything: what happens in minute 61 when the borrower's bank statement PDF is a photo of a photo?
Remind me why we moved on from FICO plus a spreadsheet? Worked fine in 2019 and nobody had to explain a neural net to a regulator.