Kensho

Inspiration

SMEs represent the backbone of most economies, yet accessing credit remains one of their biggest challenges. A major reason is that SME financial data is inherently fragmented and unstructured. Unlike large corporations, most small businesses do not maintain audited financial statements or standardized financial reporting.

While working in SME lending environments, I saw firsthand how lenders spend significant time manually reviewing documents, verifying compliance information, and reconciling financial data across multiple sources before making a credit decision.

This experience inspired me to build Kensho, a system designed to structure and connect these data points so lenders can make faster and safer decisions.

What it does

Kensho is an AI-powered SME lending operating system that connects compliance documents, business identifiers, and financial information into a unified risk view.

Instead of treating each input separately (documents, financial statements, business registration data), Kensho analyzes them together to identify inconsistencies and surface potential risk signals.

Key capabilities include:

-Document analysis and validation

-Business and ownership research

-Financial categorization and analysis

-Cross-document consistency checks

-Portfolio-level insights for lenders

The goal is to transform fragmented SME data into decision-ready intelligence

How we built it

The project combines modern AI models with structured data pipelines.

At a high level, the system works as follows:

Documents and financial records are ingested and converted into structured data.

AI models extract key entities such as company names, licenses, and ownership information.

Financial transactions are categorized and analyzed.

A validation layer cross-checks information across sources to detect inconsistencies.

The results are presented as a structured risk profile for lenders.

In simplified form:

Risk Score=f(Ddocuments,Ffinancials,Bbusiness signals)

Where each data source contributes signals used to build a holistic assessment

Challenges we ran into

One of the main challenges was dealing with unstructured and inconsistent data formats. SME documents vary widely in structure, language, and quality, which required designing flexible extraction and validation workflows.

Another challenge was building systems that not only extract information, but also connect the dots between multiple sources to detect inconsistencies.

Accomplishments that we're proud of ## What we learned

Through this project, I learned how to:

-Design AI workflows that combine structured and unstructured data

-Build validation layers that improve decision reliability

-Integrate large language models into real financial workflows

Most importantly, I learned that the real challenge in financial AI is not just extracting data — it is building systems that understand relationships between data sources and translate them into actionable decisions.

What's next for Kensho

SME finance ecosystem, connecting brokers and smes to lenders

Built With

• ai
• ai-based
• apis
• categorization
• cloud
• deepseek
• document
• engineering
• financial
• gemini
• google
• infrastructure
• next.js
• pipelines
• postgresql
• processing
• prompt
• python
• react
• rest
• vertex
• workflows