We’re building DataRiver, a tool that converts PDF bank statements into structured, Excel-ready data (transactions, balances, dates), with a strong emphasis on privacy.
Unlike generic PDF extractors, DataRiver runs on a privately developed AI model trained specifically on bank statement layouts. Uploaded files are processed by our own model and are not sent to third-party LLM APIs, nor reused for training.
This approach came from working with sensitive financial data where accuracy and data isolation matter more than general-purpose flexibility. Many real-world statements include inconsistent tables, wrapped descriptions, or mixed formats that break rule-based tools.
Key points:
Private, domain-specific model for financial statements
High accuracy on messy, multi-column PDFs
No reuse of customer data for training
Files processed transiently and not stored long-term
DataRiver is used by accountants, auditors, and founders who need clean data without risking data exposure.
We’d appreciate feedback from anyone dealing with financial PDFs, especially on edge cases, privacy expectations, or compliance requirements.
A few easy points to improve the design (and the conversion)
- Do not use the current pairing of yellow and green. It reminds me of radioactive colors. Not good for food
- Use a serif from Google Fonts to bring a premium feel, try Crimson or Merriweather for example
- Some of your gradients contain dead zone, search for clean gradient generation tool. This one
https://www.learnui.design/tools/gradient-generator.html
- Avoid white text on white. Even with the drop shadow, it doesnt work. Check for WGSA contrast color tools if you are stuck.
Hey, this is a super common problem founders face.
And you will encounter them again again even during your GTM phase.
The team at Wiz (acquired by Google for USD 32 billion) has the perfect strategy to overcome the sugar-coated feedback when pitching. They called it the "false positive"