A MENA AI studio bringing visual search to interior retail
- Sector
- Commerce
- Work
- Applied AI
- Duration
- 6 weeks
Photo in, shortlist out
no product name, category or trade vocabulary required
It stops
hand it something that is not an interior item and it says so rather than inventing furniture
Once per search
the photograph is read once, not again on every page of results
No account
the shortlist lives in the browser until it is exported as a PDF
Fig. 1
Search words instead of a visual index
Rather than build and maintain a visual index of the catalogue, the product turns the photo into search words and uses an existing shopping index.
- 1. The usual route means taking in a whole catalogue, turning every product picture into comparable numbers, indexing them and keeping it all in sync.
- 2. Instead, the photograph is described in the words a shopper would type.
- 3. Those words go to an existing shopping index, maintained by others, which finds the products.
01The starting point
Customers could see what they wanted, but not describe it.
Someone sees a chair in a photograph and wants to know what they can buy that looks like it. There is no obvious way to ask that question. You cannot search with a picture, and describing what you are looking at (the style, the material, the shape) takes exactly the vocabulary that someone outside the trade does not have.
Matching a photograph against products by visual similarity is a genuinely hard problem, and an expensive one to get wrong. A system that returns confident, plausible, incorrect matches is worse than one that returns nothing, because nobody can tell the difference until they have wasted an afternoon.
Fig. 2
A photo, but no words to search with
Customers could show what they wanted but could not describe it, and every search box starts with words.
- 1. The customer can show what they want in a photograph.
- 2. A search box asks for a product name, a category or the trade’s word for it, and they have none of these.
- 3. The products are there, but only the right words lead to them.
02What we built
A practical route from image to product.
A practical route from image to product. It goes to a vision model asked for one thing only: the words you would have typed into a shopping search if you had known them: style, material, colour, item type. Those words then go to a shopping index that already exists and is already maintained, pinned to the UAE so the results are things you can actually get here.
That decision is the whole product. It has one cost: if the image is outside the intended category, the model can still produce plausible product words. A category check was added so the product says the image is unsupported instead.
Paging through results reuses the words already drawn from the photograph rather than reading it again, and there is no account of any kind: the shortlist lives in the browser until it is exported as a PDF.
Fig. 3
Turning a photo into a search
The image is described in the words a shopper might type, and those words search an existing product index that returns real products.
- 1. The photograph is turned into a few search words, such as “rattan” or “low armchair”.
- 2. The words search a shopping index that already exists, held to what is on sale in the UAE.
- 3. Real products come back, each with its price and the shop that sells it.
- 4. A result with no seller attached is left out.
03How the product works
From a photograph to something available in the UAE.
Fig. 4
Rejecting photos that are not interiors
The model will describe furniture in any photo, so a check stops the search when the image is not of an interior item.
- 1. Someone uploads a photograph of something that is not an interior item, such as a person or a pet.
- 2. The model is told to say so when the picture is not of an interior item.
- 3. Without that check it would invent furniture words anyway, and the search would return real products that have nothing to do with the picture.
- 4. Instead, no search runs and the customer is asked to try a different image.
Fig. 5
A shortlist to share as a PDF
Customers keep the products they want, set quantities and export the list as a PDF, with no account and nothing stored on a server.
- 1. Keep the products you want and set how many of each.
- 2. There is no account and no copy on a server: the list stays in your own browser.
- 3. Export the list as a PDF to pass on.
04What became possible
A product shortlist without the right words.
From a photograph to a shortlist you can send
Photo in, shortlist out
A person searching → One upload
Reading the photograph, when paging through results
Once per search
Once per search, not once per page
Client timeout, once the two-hop chain proved slower than assumed
Four times the patience
15s → 60s
Cost ceiling on a route anyone can call
5 per 5 seconds, 20 per minute
None → Stacked burst and minute windows
Built with
Vue 3PiniaFastAPIgpt-4o-miniSerpAPIWeasyPrintJinja2RedisDocker
Have a product idea in mind?
Bring us the product, workflow or customer need you are starting with. We will help work out what it needs to make possible.
Keep reading.
Staffing & Recruitment
From Incoming CVs to Client-Safe Candidate Profiles
Manlytic Group’s recruiters work from their phones. CVs reach them on WhatsApp as scans, photos and two-column tables, passport details have to be exact, and anything sent to an employer has to leave the agency as the only route to the candidate.
- Every file made for an employer is built without the candidate’s contact details, then searched again before it can be shared.
- CVs are read without guessing: gaps stay visible, and anything unsure waits for a person to confirm.
- A likely duplicate is reported and held for a recruiter, never merged on its own.
Research & Knowledge
From Printed Archive to Research System
The collection existed. Answering a question about it did not. Anything worth knowing meant reading across thousands of pages by hand, holding names, places and dates in your head, and reconstructing the evidence trail yourself each time.
- Ask a research question and get back a cited report, instead of reading for the answer
- Follow one person across the whole collection even where their name is spelled a dozen different ways
- See the archive as a map of what happened where, a timeline of when, and a network of who dealt with whom
Scholarly Publishing
Publishing Platform for Four Peer-Reviewed Journals
How a scholarly publisher brought peer review, editorial decisions, DOI registration, preservation and publication into one bilingual platform.
- One manuscript record carries the whole process (submission, review, decision, issue, publication) instead of being reassembled from email
- Each person sees the view that belongs to them: authors follow their own submission, reviewers open only what they were assigned, editors see the operational state
- Publication is one operation rather than a checklist: the article, the DOI deposit, the preservation copy and the author notification are coordinated together