Hi, I'm Kellan

I spent years building integration and AI automation. I was a solution engineer at Make, helping companies wire together APIs and workflows. Now I lead business automation and AI work in my day job. Most of that time has been spent solving the same problem in different shapes: data arrives in one format, the system expects another, and someone ends up retyping it.

One pattern kept showing up: parsers are good at reading text off documents—PDFs, scanned images, Excel files—and they can be useful pieces of a build. But the gap isn't in the extraction. The gap is in what happens after: matching a customer's part number to your SKU, converting “box of 100” to eaches, applying contract pricing, and deciding when something is certain enough to post automatically versus when it should wait for a person to review.

I built an AI invoice automation in Make for a multi-supplier retailer with 20+ suppliers. It replaced an earlier parser-based build that had failed. Fuzzy vendor matching so slight name variations didn't break the workflow. Category tax logic that applied the right rules per product type. Total validation to catch invoices where the line items didn't add up. About 200 hours a year saved. The breakthrough wasn't the AI—it was designing for the edge cases and putting a review queue in front of the posting step, so nothing unclear made it into the ERP on a guess.

Parsers handle clean documents. Your customers don't send clean documents.

MasterDesk applies the same approach to purchase orders. AI extraction handles PDFs, Excel, and email body text. Custom cross-reference tables match customer part numbers to SKUs. UOM and pricing rules get applied per line. Anything uncertain goes to a review queue. Clean orders post to Acumatica automatically; exceptions wait for a person.

I'm building this for Acumatica only, on purpose. One ERP done well, instead of six ERPs done halfway. The Acumatica REST API is stable, well-documented, and designed to be used. Fastener and MRO distributors already run on it. That focus lets me build the matching rules and validation logic instead of fighting a different API every month.

I'm in Evanston. If you're anywhere around Chicago I'd rather drive out and meet you than do this over a screen share. And when you call, you get the person who built the thing. There's nobody else here to hand you off to.