Kita bets AI can read messy loan documents; $4.5M seed, $130M processed

In the Philippines a loan application often arrives as a blurry rural bank form, a handwritten passbook, or screenshots of GCash transfers — evidence of creditworthiness buried in documents that legacy OCR cannot read. Stanford classmates Carmel Limcaoco and Rhea Malhotra founded Kita in 2025 to make that mess legible: its vision-language models read more than 50 document types, flag inconsistencies, apply each lender's credit rules, and draft a credit memo with every figure linked to its source.

Kita is not a lender. Its 'AI credit officer' works behind the scenes for banks, fintechs and microfinance institutions — requesting missing documents, chasing borrowers over Viber, WhatsApp or SMS, and leaving final credit decisions to humans. The founders deliberately chose the hardest customers first: traditional banks and large microlenders holding decades of paper records.

The bet has traction: within five months Kita said it processed over $130 million in loan volume, more than 60% of it from the Philippines; for TRBank it turned four years of paper lending history into searchable data in under three days at above 97% verified accuracy. It went through Y Combinator's W26 batch — the company says Limcaoco is the first Filipina founder admitted in over five years — and in August 2026 closed a $4.5M seed led by BoxGroup.

Kita is live with lenders in the Philippines, Indonesia, Mexico and the US, and pitches its traceability — every figure mapped back to its source document — as the reason it differs from both legacy OCR and other AI underwriting tools. Its name comes from the Tagalog word for 'to see' and 'earnings'.