How the AI interior design platform works

VITADOMUS is built to behave like an interior architect, spatial planner, product researcher and purchasing assistant in one connected platform. It does not simply take a photo of a room and paint a prettier one.

The pipeline (tap each step)

Why several services, not one model

The customer may start with a single sentence, or upload images, plans, measurements and inspiration. VITADOMUS converts whatever arrives into a structured design brief.

“I want to redecorate my living room. It feels dark and cluttered. I like modern interiors, but I don’t know what will look good. I have about £3,000 to spend.”

What the AI extracts from conversation

Reasoning from lifestyle

“We are a family of four. We watch films in the evening, entertain friends at weekends and need somewhere the children can play. We prefer a calm interior, with little maintenance.”

• Comfortable viewing positions and sufficient seating.
• Circulation that stays usable when guests are present.
• Concealed or easily accessible toy storage.
• Materials suitable for frequent use.
• Adjustable lighting for different activities.
• Furniture arrangements that avoid unnecessary obstruction.

It also recognises trade-offs. A large sectional improves seating but reduces play space and circulation. Rather than choosing silently, VITADOMUS presents two alternatives and explains the compromise.

VITADOMUS needs to understand not just what a room looks like, but what it physically contains and where objects can go.

Understanding different inputs

Constructing the room model

For a rectangular room the system may receive: length 4,000 mm · width 3,000 mm · ceiling 2,400 mm · door on the front wall · window on the right wall · radiator under the window · existing sofa and television. It turns these into objects with coordinates, dimensions and relationships.

Provenance: where did each number come from?

Every measurement carries a status and a source. Even customer-measured dimensions stay subject to the customer’s confirmation before anything is bought.

Asking only what matters

“You have provided the room dimensions and window location, but I still need the clear door opening width to assess furniture delivery. Can you measure this?”

The spatial engine asks only questions relevant to the decision in hand, and the customer can correct dimensions directly on the floor plan.

Once the brief and room model exist, the AI translates them into design possibilities. This is where VITADOMUS should behave more like a designer than an image generator.

4.1 Establish a hierarchy of constraints

4.2 Develop and test alternative arrangements

A spatial-planning engine generates arrangements and tests each one against the project objectives and physical requirements. For a living room it might compare:

• Sofa opposite the television.
• Sofa against the longer wall to increase circulation.
• Two smaller seating pieces instead of one large sofa.
• A multifunctional storage-and-media wall.
• A flexible arrangement that supports different household activities.

4.3 Apply interior design knowledge

Not every principle is a legal or technical standard. VITADOMUS should distinguish four classes: binding requirements, recognised technical guidance, professional design practice and subjective taste, and label advice accordingly.

4.4 Produce an explainable proposal

“I recommend positioning the sofa along the longer wall because this preserves more central floor space, keeps a clearer route to the door and leaves room for a reading chair. I have selected a warm neutral palette to support your calm atmosphere.”

The exact conclusion depends on real dimensions and the customer’s priorities. What never changes is that the reason is shown, so the customer can challenge it.

VITADOMUS should never ask a language model to invent a shopping list from training data. It retrieves actual product records from authorised retailer sources. This is one of the most important parts of the business.

The retrieval flow

5.1 Where does product information come from?

A dedicated VITADOMUS Retail Intelligence Engine would use several sourcing methods, each with different permission and data quality:

Google’s merchant product-data specification defines fields such as price, availability and identifiers, but that does not give a general third-party app unrestricted access to every seller’s Merchant Center data.

5.2 How retailer selection works

“Only search established retailers delivering within the UK. I don’t want products from unverified marketplace sellers.”

The retailer policy is applied first, restricting which catalogues and sellers are eligible. Approved brands such as IKEA, John Lewis or Dunelm would be included only where suitable authorised data exists; a retailer’s reputation alone does not establish a live integration. If the customer instead allows independent sellers, the system assesses seller identity, authenticity evidence, returns and delivery policies and reputation signals. The customer stays in control of which suppliers are acceptable.

5.3 Understanding an uploaded product photograph

If the customer uploads a Pinterest image of a green curved sofa, computer vision extracts a specification:

The image is turned into a numerical representation (an image embedding) so similar catalogue photos can be found even when retailers use different words. A text search finds “curved green velvet three-seat sofa”; a visual search may also surface a “moss upholstered crescent couch”. Crucially, visual similarity is only the first stage: a lookalike can still be the wrong size, out of stock, too expensive or impossible to deliver.

5.4 Hard filters versus ranking preferences

5.5 What a final product record contains

5.6 Keeping information current

• Import permitted product data regularly.
• Normalise and index the records.
• Detect changed prices, availability and variants.
• Refresh shortlisted products before showing purchase recommendations.
• Recheck selected products when the customer proceeds to buy.
• Warn if a selected product has vanished or become more expensive.

An affiliate feed might refresh daily or faster depending on the supplier’s terms; other APIs allow checks nearer purchase time. Only the retailer’s live checkout can ultimately establish the real order price, availability and deliverability.

Finding a suitable sofa is only part of the problem. The sofa, rug, table, lighting, storage, paint and finishes must work together as one complete design.

Illustrative product weighting

S = Σ ( wᵢ × qᵢ )

S is the overall suitability score, wᵢ the importance of each criterion and qᵢ the candidate’s measured or assessed performance on it. The customer can raise the importance of cost, durability, appearance or delivery speed. These weights are illustrative, not a validated industry standard.

Try it: scoring versus hard constraints

Relationships the optimiser also checks

• Does the rug fit under the seating arrangement?
• Do furniture heights and proportions work together?
• Are the selected colours compatible?
• Does the flooring suit the expected wear?
• Does the lighting support the room’s activities?
• Are materials and finishes consistent?
• Does the whole selection remain within budget?

This makes VITADOMUS more than a product recommender: it becomes a whole-room design coordinator.

This should be one of VITADOMUS’s most distinctive technical functions: proving the item fits the room and can physically get there.

Try it: test a delivery route

A sophisticated version would simulate the movement and rotation of a 3D package through the route. A basic width comparison is a useful first check but is not enough for corners or staircases.

For each selected item the system determines quantity, unit price and the extra costs around it, then assembles the whole project total.

For materials sold by area (flooring, tiles) it estimates quantities from the room model and applies allowances for cuts, waste, pack sizes and installation requirements.

Illustrative £3,000 living-room allocation

Example allocations only, not a quotation or a list of sourced products. Real projects use verified retailer prices, delivery charges, relevant taxes and supported installation estimates.

• If the customer’s chosen sofa costs £300 more than the previous pick, the system recalculates the project total and shows the effect on the remaining allowance.
• It may suggest a lower-cost table or rug, but only if those items are not locked by the customer.
• If the scheme cannot satisfy both budget and mandatory requirements, it explains the conflict and asks which constraint the customer is prepared to change.

Once a feasible product shortlist exists, VITADOMUS creates the coordinated interior presentation.

The visualisation engine should use retailer-provided 3D models where licensed and available. If none exist, geometry can be reconstructed from dimensions and approved product imagery, with the limitation disclosed.

Three honesty levels for every image

VITADOMUS needs a design dependency system. A change must not regenerate the whole room arbitrarily.

“I don’t like this sofa anymore. Find a darker green one, but keep my rug and stay within my budget.”

The project stores relationships between objects. Changing the sofa may affect rug size, coffee-table clearance, viewing distance, lighting and the colour palette; it must not automatically change the kitchen floor tiles or another unrelated room.

A hypothetical customer: Redesign my living room.

“My living room is 4 m by 3 m. I live with my partner and our dog. We watch films and sometimes work from the sofa. I want a contemporary, calm interior with a green sofa, new flooring and better storage. Budget £3,000. We live in the UK and want the room completed within four weeks.”

When nothing qualifies

“I haven’t found a verified product that satisfies all your requirements. You can widen the retailer selection, increase the budget, choose a smaller sofa or extend the delivery deadline.”

VITADOMUS is designed around a set of interconnected services that all read and write the same project record.

A shared source of truth

A product must not exist only as text in a chat transcript. It needs a unique identity linked to its room, position, quantity, retailer, design version and supporting evidence.

This makes the project editable, auditable and reusable.

“I want to replace my living room rug. Keep everything else, find something more durable and spend no more than £250.”

Six months later VITADOMUS can retrieve the approved arrangement, product details, palette and preferences, then search for currently available compatible replacements, without rebuilding the project from memory.

The platform is given several levels of authority. The more irreversible the action, the more explicit the customer’s consent must be.

The full ambition is technically substantial, so development is separated into practical levels.

A few hundred well-structured products, adequate for the initial design categories, may be more useful than millions of poorly described listings with missing dimensions. The required catalogue size is to be established in testing.

The long-term objective is a living digital interior model that customers can continually adapt. Every room, product, material, design decision and approved modification becomes part of an evolving property record.

From understanding your life to designing your real home. The customer describes their needs, the platform understands their space, AI develops appropriate designs, real retailers supply the products, and structured checks help make the project achievable.

The operating model describes what VITADOMUS does. A production system also needs the layer below: how it knows it is right, what it does when it is not, and how it improves.

The Vision stages describe the customer experience very well. The operating model adds the rules, evidence and safeguards underneath it. This table checks one against the other.

Recommended additions to the Vision