Platform
Seven layers. One flow of information.
Data captured at the till reaches the forecast without a single integration, an export or a nightly CSV. Here is the whole stack, top to bottom.
7
layers, from capture to automation
12
operational modules on one schema
11
models in production, retrained weekly
0
integrations needed between them
Architecture
Read it bottom-up: capture, then meaning, then action.
Select a layer for what runs there, what it owns, and what it hands upward.
Information travels upward continuously — and actions come back down.
Intelligence
Models find the patterns: margin carriers, customer segments, anomalies, menu performance.
Models in production11
Retrain cadenceweekly
Features per outlet180+
Validation window4 weeks
Hands upward
Features and learned patterns the forecasts are built on.
One event, seven writes
Follow a single order through the stack.
Order #1284, two chicken biryanis, table six, 19:42. Nothing is entered twice and nothing is inferred later.
Elapsed from tap to forecast update: under 2 seconds
DataOrder event written once, with outlet and table19:42:03
OperationsKOT fires to the hot station; table six marked seated19:42:04
OperationsRecipe deducts 620g rice and 480g chicken from stock19:42:04
AnalyticsItem mix, daypart and channel aggregates updated19:42:05
IntelligenceCustomer record recalculated — Rhea M. moves to Regular19:42:05
PredictionChicken depletion time revised to 20:10 tonight19:42:05
AutomationReorder draft prepared for Vendor B, pending approval19:42:06
Engineering
Built for a restaurant, not a data centre.
Offline-tolerant billing
The till keeps taking orders when the link drops and reconciles on reconnect.
Queued writes replay in order
Single write path
Every event has one authoritative write, so numbers never disagree between screens.
No nightly CSV, ever
Outlet isolation
Roles and data scoped per outlet, with a group roll-up above them.
Enforced at the query layer
Fast at peak
Billing and KDS screens are budgeted for the Friday rush, not the Tuesday demo.
Billing screen under 400ms
Auditable history
Voids, discounts, price changes and overrides keep who, when and why.
Immutable audit trail
Models on your data only
Your history trains your models; nothing is pooled across restaurants.
Per-tenant training
Walk the stack with your own data in it.
We'll trace one of your real orders from the till to the forecast on the call.