FinCleanse

When I started building FinCleanse - a transaction enrichment API - I assumed most of my time would be spent writing code or creating new features. Turns out that hasn't been the hardest part (not even close).

The hardest part has been everything around the code.

1) Building the API was the straightforward part

I'm a cloud engineer by background, so building the core system felt familiar and comfortable. I immediately jumped into the "build" phase when I started FinCleanse as that's the part I was most excited about and prepared the most for. I already had an architecture in mind and the stack defined (cloud-native, AWS, serverless). I was ready to get building. However after building the MVP I realized that was only a fraction of the work.

2) The majority of time goes into understanding the market

I intentionally wanted to keep FinCleanse simple before writing too many additional features that I think would be valuable for the product. I told myself that I would add these features either once I get some valid market direction towards those or if customers show interest. Instead I've kept it light and spent a surprising amount of time on:

• researching existing vendors in the transaction enrichment space

• figuring out how they position themselves

• understanding pricing models

• identifying where an API-first product fits into the landscape

Not glamorous work, but probably the highest leverage work I've done so far.

It's easy to build something but it's it's harder to build something that fits into a real gap.

3) Marketing is a bigger engineering problem than I expected

This one caught me off guard. In terms of both the tools and the thinking mentality.

Questions like:

• Who exactly is this for?

• What message actually resonates?

• Should I position this as enrichment, cleansing, normalization, or analytics?

• Who do I reach out to first?

Writing outreach emails has taken more iteration than writing parts of the API and unlike code, there's no test suite that tells you when your messaging is wrong.

The tooling aspect has been surprising as well. Up until this project I haven't really messed around with any sort of outbound products such as Apollo. But since using those for this project, the amount of time I've spent learning outbound, sequencing, domain ramp up, etc. has surpassed way more than the time I spent initially creating the MVP. Honestly it feels way more like an art that I ever expected it to be.

4) Early-stage building feels more like research than development

Right now, building FinCleanse doesn't feel like "shipping features."

It feels like:

• validating assumptions

• narrowing scope

• figuring out where the product actually creates value

• figuring out what functionality is sticky and resonates with people

The coding is still there but it's not the center of gravity and hasn't been for the past few weeks. Understanding the problem space is and figuring out how to efficiently put the product out there and get feedback has become the focus of my attention.

What's surprised you the most when building your first product?

Was it technical? Or was it everything around the product itself?

If you've ever worked with raw bank transaction data, you know the problem. A charge shows up as "SQ *COFFEE SHOP 84920 CA" and your app has to figure out what to do with it. Is it food? Is it local? What's the merchant called? Is there a logo?

The existing solutions are either locked behind enterprise pricing, bundled into platforms you don't need, or just not built with developers in mind.

I'm a cloud engineer who started digging into this space and realized that having access to this type of clean data can help accelerate numerous use cases in the fintech industry - customer PFM, CU/bank cross-sell analytics, spending insights, and more. So I decided to build an API that any product could plug into for its own needs.

FinCleanse is a transaction enrichment API that takes raw transaction strings and returns structured merchant data: name, category, and logo. The core enrichment layer is built and live. I just launched a public demo at demo.fincleanse.com where you can try it with a natural language search interface and see what it returns.

This is early. I'm not here to pretend otherwise. What I'm looking for right now is feedback from developers who've dealt with this problem, fintech builders who've evaluated existing options, or anyone who takes the demo for a spin and has thoughts.

If that's you, I'd love to hear from you.

Thank you!