Kita, a Y Combinator W26 startup, is tackling the broken credit infrastructure in emerging markets with AI. Founded by Carmel and Rhea, the company uses vision language models (VLMs) to automate document review for lenders in places like the Philippines and Mexico where traditional credit bureaus are unreliable.

Existing OCR and document AI tools fail on the messy, unstandardized financial documents common in these markets. Lenders typically rely on borrowers submitting bank statements, payslips, and other documents in various formats - PDFs, images, screenshots - making manual review slow, expensive, and error-prone. The founders discovered this firsthand while testing every available tool, finding none could handle real-world document chaos.

The scale of the problem is massive: in 2025, $13.3 trillion was lent globally, with 90% of transactions involving document review. Kita's solution parses 50+ document types, detects fraud, and extracts underwriting signals using a model-agnostic VLM architecture. The system improves by linking document signals to repayment outcomes and training on localized credit data from each market.