Real volume. Real businesses. Real production.
Most of what we build sits inside operations that aren't public-facing. The names stay private. The numbers, the scope, and the constraints don't.
AI pipeline shipped into a production ERP.
ScopeArchitect and ship the full AI layer inside an existing ERP serving 70+ companies.
ContextWe didn't build the ERP. The team behind it brought us in because their customers needed an AI layer that could actually run on real data — not a chat bolted on top. They trusted us to architect, build, and ship it end-to-end.
of customer operations now flow through the layer.
A real ingestion pipeline that prepares each customer's production data for retrieval — on the actual shape of their business, not a generic schema. Edge cases and inconsistencies handled in code, not pushed onto the LLM.
Retrieval architectures shaped around the ERP's data model. Vectorization tuned per data class. Queries reconciled against the source of truth before generation, not after.
A caching layer so the system doesn't pay for the same thought twice. Vectorization batched and persisted to keep cold-start latency predictable under load.
Outputs that touch finance, operations, or stock are verified against the underlying numbers programmatically before they leave the system. The LLM proposes; the system checks.
Managers ask the system for a custom report, in natural language, on their own data. The system produces it — formatted, validated, reconciled. In production, per company.
The result is an automated manager that 70+ companies — and $5M+/week of their operations — actually trust to run. Not an API wrapper. A system.
The complete backbone for a perfume retailer.
ScopeDesign and ship the entire operational stack from scratch — internal ERP, custom POS, customer ordering app.
ContextA specialty retailer needed a single coherent system. Off-the-shelf POS, off-the-shelf ERP, off-the-shelf ordering — none of them spoke to each other, and none fit the actual flow of selling perfume. We built one system with three interfaces instead.
in operations on the system today.
Inventory, suppliers, purchase orders, stock movement, employee management. One backend, one schema. Built for the staff who use it daily, not for the demo.
Designed around the actual choreography of a perfume sale — sampling, multi-item, gift wrapping, customer recognition — not a generic retail script.
Customers order, staff fulfills, stock decrements in real time. Same source of truth. No reconciliation jobs running overnight.
One backend. Three interfaces. One source of truth. The point wasn't shipping three apps — it was making one system that behaves like one system, from the warehouse to the till to the customer's phone.
Restaurant ERP across two countries, since 2020.
ScopeEnd-to-end operational system for a hospitality group — multi-platform, multi-modal, offline-first, owner-online.
ContextLive in 4 restaurants across two countries with unreliable connectivity. That constraint shaped every architectural choice. The system runs whether the internet does or doesn't.
since launch in 2020. 4 establishments still running it today.
A small headless mini-PC under Linux acts as the local server in each establishment, holding the source of truth on the local network. Survives outages. Runs the floor.
Servers take orders on phones, send them straight to the kitchen and bar. Multi-device, multi-floor, all syncing locally to the central POS.
Tickets print to the right destination automatically — caisse, bar, kitchen — based on the product. No double-handling. No staff routing tickets manually at peak hours.
Stock decrements on every order, across every device. End-of-night reconciliation isn't a job; it's already done.
Everything runs on the local network, with or without internet. Orders, payments, prints. Connectivity is a sync mechanism, not a runtime requirement.
A robust online layer batches local data to the cloud on a regular cadence. Owners watch revenue, recipes, and performance across all their establishments in real time when they're online.
One of the systems we're most proud of — not because it's the biggest, but because it's still running. Five years in. Zero churn. In the conditions it was built for.
Smaller systems. Same principle.
Beyond the headline projects, we've shipped a number of smaller operational systems for SMBs — internal dashboards, custom workflows, integrations between tools that didn't talk to each other. Different scopes. Same principle: software that runs the business, not software that sits next to it.
Long-term digital systems management.
Active retainers with several companies — web presence, internal tools, monthly evolutions. 4+ production websites shipped across retail and hospitality along the way.
The kind of relationship where you don't have to re-explain the business every quarter.
Got something similar in your operations?