Most of the shops we work with run on an ERP that does one thing well, and spreadsheets for everything around it. We build the part in between: quoting, travelers, the floor kiosk, the order that gets retyped twice.
They build warehouse automation and sell it worldwide, and a buyer in the States had nowhere to land. The corporate site speaks to a different market in a different language. The US team had no address to send anyone to, and no way to answer what a system costs here without going back to head office and waiting.
What we built
A US site with its own pages for end customers and for distributors, because the two want different things and were being handed the same brochure. Behind it sit three tools the sales team actually uses: a site survey they fill in while walking a warehouse, a system sizing estimator, and a landed cost model that turns a factory price into a US one.
Where it stands
The sizing estimator is the part that matters. A rep captures tote dimensions, tote count and orders per hour on a form, and comes back with a system size and a price that already carries ocean freight and import duty. That arithmetic used to live in a spreadsheet on someone else's continent. The site is built and the tools work; the lead forms and analytics are being wired before launch.
ManufacturingBuilding now
Custom marketing website on own domain
For a manufacturing company
The problem
They convert specialty materials and run warehouses across the country, and their web presence could not answer either question a buyer actually asks: what can you cut, and how close are you to me. Product data, capability specs and locations were not in any system that could render them.
What we built
A site where the content is a database rather than markup. Eleven sections, all of it in tables, so a new product or a new warehouse is a row and not a rebuild. The hero is a mock cut ticket carrying material, width, cut method, cores and lead time, because a photograph of a factory floor tells a buyer nothing a competitor's photograph does not.
Where it stands
The US map is generated rather than licensed. An Albers projection was rendered to SVG once, and each warehouse stores its own projected coordinates, so opening a new one is an insert. It went in seeded with their real catalogue: 25 products, 18 capability rows across six groups, seven warehouses. The admin panel that lets them edit all of it themselves is the next piece.
ManufacturingBuilding now
Shop-Floor Job & Quoting Platform
For a manufacturing and repair shop with a walk-in counter
The problem
Every motor that arrived was written into a physical book and a paper tag was tied to the housing. Work got picked up off the floor when someone felt like it, so nothing ran in order. The shelves were the real problem: stripped motors stacked up with no way to tell what was waiting on a part, a quote or a decision. In the owner's words, you look at it and you think, what is this again.
What we built
An intake queue that sorts arrivals into a pool, into assignments, and into a quote gate for the question that decides everything: is this motor worth repairing at all. A six stage job board. Five department queues, so when a tech writes down what they need it lands in purchasing, machining, fabrication, rewind or an outside vendor rather than in a conversation.
Where it stands
The decision that made it work was refusing to build a second system. A logged motor and a work order are the same physical thing, so the new workflow record and the existing repair record bind to each other the first time a job is opened. Years of history stayed exactly where it was. Every migration is additive, nothing was rewritten, and the shelf is now a board.
ManufacturingBuilding now
ERP Workflow Automation for Manufacturer
For a custom manufacturer running a legacy ERP for quoting and order processing
The problem
Two clerks sat in front of two screens each: email and customer PDFs on one, a Windows desktop ERP on the other. Every purchase order was retyped by hand between them. Part numbers could not be pasted, so each line meant opening a search, finding the part, and matching the revision letter by eye. Finding the right customer record was its own hunt.
What we built
Drop a purchase order in and it comes back parsed, with a confidence mark on every field and the whole thing checked against the live catalogue: the part exists, the revision matches, the quantity lands on the right price break, the requested date works against transit time. Anything wrong is fixed inline, then pushed into the ERP.
Where it stands
The finding that redirected the project: saving one order in the desktop client writes seven tables, and a plain insert of the header and lines writes four. The missing three come from the desktop application itself, not from the database, so the web app has to reproduce them. That was established by holding a snapshot while somebody clicked Save. Writes are proven against a sandbox, not production.
ManufacturingBuilding now
Shop-Floor Execution System for Manufacturers
For a software venture serving small and mid-market manufacturers
The problem
Routes, production schedule, capacity and manpower all lived in spreadsheets, and the ERP was used for exactly one thing: opening work orders. On the floor, a paper traveler walked with the parts. A real one from a single run carries six lot numbers written in four incompatible formats, and one component line with two lots on it because a roll ran out partway through.
What we built
Work orders with real routing, scheduling, and a floor kiosk where people clock in, take the next job and scan lots and serials at the station. The traveler becomes an issued record: drafted, issued, then closed against a frozen snapshot. In-process quality checks block a job from ending until they are answered, and a step needing two signatures will not accept the same person twice.
Where it stands
The scheduler proposes and a human always authorises; nothing auto-applies to a floor where people are standing. The connection to the existing ERP is read-only by default and guarded in code, and costing is advisory, because this system is explicitly not allowed to become the book of record. A defect found in pilot is the kind of thing that only shows on real data: nothing ever marked an operation done, so on seeded data the flow walked and on theirs the second step would never have appeared.
Vibration AnalysisDelivered
Adopted an off shore code base
For Failure Mode Detection Company
Adopted a 3 year old codebase that was written offshore
Cleaned code and brought project to a viable state
