Architecture
One decision engine. Instantiated per domain.
Every OKAO product is an instantiation of the same loop. What differs is the domain data, the constraint model, and the shape of the decision at the end of it.
The loop
- 01
Intent
What is actually being attempted
- 02
Context & Data
Domain signal, constraints, history
- 03
Inference
Models fitted to the domain
- 04
Recommendation
Ranked, explainable options
- 05
Decision
A human commits
- 06
Outcome
What the world returned
- 07↺ 01
Learning
Priors updated, loop closes
- THE LOOP
Every domain we enter reuses this spine. Only the data, the constraints, and the shape of the decision change.
Same layer, different world
| Layer | Shared mechanism | Resactly | Energy |
|---|---|---|---|
| Intent capture | A typed representation of goals, constraints, and tolerances. | Occasion, party, budget, distance, dietary constraints, appetite for novelty. | Load profile, siting envelope, reliability target, capital limits, timeline. |
| Domain data layer | Entity graphs, normalization, and provenance per industry. | Restaurants, menus, dishes, preparation, service patterns, behavioral signal. | Generation, grid and interconnection, geography, fuel and market pricing. |
| Inference | Models fitted to domain structure, scored against realized outcomes. | Dish-level affinity and match scoring under contextual conditions. | Feasibility, cost-to-serve, and risk evaluation across configurations. |
| Recommendation surface | Ranked, explainable options with the reasoning exposed. | A short list a diner trusts, with the reason it fits stated plainly. | Comparable configurations with sensitivities and stated assumptions. |
| Outcome & learning | The decision and its result return to the system as signal. | Visits, orders, repeat behavior — and demand intelligence for operators. | Project progression, revised estimates, and post-decision reality. |
Why this stays focused
A new domain is only worth entering when four conditions hold: intent is rich and currently unrecorded, the data is specialized enough that generic tools fail, a real decision sits at the end of the process, and the outcome of that decision is observable. Most industries fail at least one. The two we work in do not.
Intent is rich and unrecorded
Data is specialized and messy
A real decision terminates the process
Outcomes come back measurable