Selected work
What we've built
Most of our work is under NDA, so the clients below are described rather than named. The numbers are not — every figure is one we measured ourselves on the engagement it sits under.
Advisory Firm Data Assessment
SEC-registered RIA — confidential
Finding out what the AI project actually was, before building it
The firm wanted AI-assisted client summaries. Before writing anything we went through every contact record in their CRM — read-only, nothing changed, no client data stored — and found the real problem: an advisor was assigned on 9.5% of client records, and "advisor" was recorded across four different fields depending on who had entered it. Any automation built on that would have produced confident, wrong output. What we delivered was a plan the firm could act on, in the order that made each next fix possible. The Wealthbox-to-Outlook sync that came out of it was a byproduct, not the point.
Read the case study →Storefront Performance Investigation
Top-10 antique agriculture retailer — confidential
Three days of random error pages, traced to a limit the platform never published
A merchant’s storefront started failing on roughly half of all page loads, on every kind of page, with nothing in the theme settings to explain it. We wrote a small script that loaded the storefront fifty times and recorded what came back, how long it took, and a number Shopify quietly attaches to every page saying how much work that page cost it. The cause was a handful of lines in our own add-on that re-read the store’s saved settings on every single page, pushing each one past a cost limit Shopify does not publish anywhere. We published the entire investigation, including the four hypotheses that turned out to be wrong and the probe script itself, so any merchant can measure their own store.
Read the case study →Dashboard Platform Re-architecture
Nationally recognised brand — confidential
One database doing three jobs that got in each other’s way
Three jobs were sharing one database and pulling against each other. The dashboards had to answer straight away for thousands of people checking them through the day. The reports had to grind through far more data than any dashboard touches. The overnight imports had to load large batches without either of the other two noticing. Something had to give, and it was always the dashboards — they are the ones with people sitting in front of them, waiting. So we split the data by how it gets read rather than by what it is. The dashboards now read from a second copy, kept in the shape they actually ask for, which turns a lookup into something closer to opening the right drawer than searching the whole building. The original records stay where they were and handle the reports and the everyday writing. Filtering was the other half of it: rather than write fresh code for every combination of filters someone might pick, one piece assembles the query from whatever they choose. From the outside nothing changed — the app asks for what it needs and never finds out there are now two places the answer can come from.
Read the case study →Reporting & Client Data Analysis at Scale
Confidential engagements
Dashboards over tens of millions of records, and AI that knows when to ask a person
Reporting and analysis platforms working across tens of millions of records. At that size, how the data is arranged and asked for stops being a matter of taste and starts deciding whether a dashboard loads at all. We also put an AI model to work on the step where raw information gets pulled in and turned into something usable — on a short leash. It is allowed to decide for itself where a wrong answer is cheap to correct, and has to hand the call to a person wherever being wrong would be expensive. That brought in 60% more information without pushing the error rate the wrong way.
Read the case study →ViewForge
Shopify-native YMM & fitment search across 8 verticals
Our own product. Cascading Year/Make/Model dropdowns, a fitment badge on the product page, and a saved-vehicle garage — with fitment stored as native Shopify metaobjects rather than in our database. That is the design decision the whole app turns on: your fitment data stays part of your store, readable by your theme, and stays yours if you uninstall. Smart Parse lifts fitment out of product copy you already wrote, eight verticals ship built in, and Premium adds VIN decoding and ACES/PIES import.
Visit ViewForge →