What should we develop next?
Ripple — product R&D register for manufacturers
Half our first ideas were already on their shelf.
What it is
A product-development register for a bottled-water R&D team. The first draft proposed alkaline water, sparkling cans and glass bottles — then we scraped their real catalogue and found all thirty-eight of them already existed. Every concept was rebuilt around genuine whitespace. The model never does arithmetic: a deterministic engine scores, and the advisor reasons over the scores.
The numbers
- Real SKUs scraped first
- 38
- Backend dependencies
- 1
- Arithmetic done by the model
- None
- Shipped vs proposed
- Photos vs CAD
What is inside
- Deterministic scoring shown beside the advisor, both labelled
- Shipped products get photographs, proposals get CAD drawings
- One engine module imported by both browser and server
- Always check the client’s catalogue before pitching them ideas
In more detail
How a new product idea gets scored
Five criteria, weighted to 100. Not already in the range carries 30. Made on lines the plant already owns carries 20, and so does whether a retailer will give it shelf. A clear regulatory path carries 15, and margin above the base SKU carries the last 15. Each criterion is answered out of five and scaled by its weight, so the same answers always return the same number.
Why the client's catalogue is scraped before any idea is written
The register starts with the shelf, not with a brainstorm. 38 live SKUs were pulled in first, across still water, alkaline, sparkling, zero-sodium, recycled PET, glass for hospitality and flavoured. Two of the five concepts on the board came from our first draft, and both were already in that catalogue. Anything already stocked scores nothing on the first criterion, so it sinks to the bottom of the ranking on its own.
What the advisor is allowed to decide
Nothing numeric. The engine computes; the model reasons over numbers it did not produce. The two panels sit side by side and each is labelled with its source, so a reader can tell which sentence came from arithmetic and which from judgement. The advisor also has to carry the unwelcome results — a concept can be genuine whitespace and still need a filling line the plant does not own.
Telling a product that exists from one that is only proposed
A photograph means the SKU shipped; a drawing means someone is proposing it. It sounds cosmetic and it is not: once a rendered concept is presented like something already on the shelf, a meeting starts arguing about the wrong thing. Concepts that turned out to exist carry a marker saying so, and the advisor's line on them is to drop them rather than rework them.
Built with
It boots in your browser and runs the real logic — no form, no sales call.
Common questions
Is Ripple a live system or a working demo?
It is a working demo, built in 2026 around a bottled-water R&D team in Abu Dhabi. The scraped range, the scoring engine and the concept board all run on this site — you can open it and click a concept to see the working — but it is not presented as software deployed inside the client's business.
Does the AI decide which product to develop next?
No. Every calculation is done by a deterministic engine, and the model only reasons over scores it did not produce. That is why the same inputs always return the same ranking, and why the working is printed beside the recommendation instead of being summarised by it.
What would it need to run on our own catalogue?
Your live range and the criteria your team actually judges ideas on. The weights are a decision, not a default — a plant with spare line capacity weighs manufacturability differently from one running flat out. Scope decides the figure: a call, then a written scope with one fixed figure and its payment stages.
Does it run in Arabic as well as English?
Yes, both, with the layout mirrored — the register, the scoring table and the advisor's reasoning all render in Arabic as well as English. Bilingual work is part of every build here rather than an option added later; we do not build the single-language kind.
More of the work
- JD PROP AIReal-estate command centre
- JD Property OSManage and maintain, one account
- Atlas ERPAI-native ERP for GCC contractors
- StockBookBuilt around one man’s spreadsheet
- ComplyGCCAML / KYC for UAE fintechs
- MRP Businessman ServicesPublish what the industry hides
- Estate VaultFor land deals that arrive on WhatsApp
- CourseDeskTutor-side automation
- ScanversePhone to photoreal 3D
- Alaska MachineryInjection moulding · Sharjah
- Aluminium & Glass demoNothing typed in from a catalogue
- Tyre & Oil demoPrices benchmarked to the real market
- Cafe demoThe licence says cold and hot
- Barbershop demoBarbering, booked properly
- Excel PlasticsEvery bottle solved, not modelled
JD Software Solutions — a four-person software studio in the United Arab Emirates. We build the system, hand over the code, and stay on to run it.