Tipografia3D: how we built Romania's first 3D printing service that measures from a photo

There's a mundane moment where people give up on a 3D printed part: the moment they're asked for an STL file. For someone who needs a shelf bracket around a pipe, a replacement knob for a washing machine, or a door stopper that fits an unusual hinge, the request ends the conversation. They don't know CAD, they have no file, and there's no way they can justify a 100 EUR per hour design fee for a 4 EUR part.

We built Tipografia3D as a direct answer to that friction. It's a 3D printing Romania platform where the customer never uploads a file. They photograph the space with a credit card as scale reference, the system extracts the measurements, they pick a parametric template and receive the part at home in 3-5 days.

Market context

We mapped 112 active 3D printing services in Romania, distributed across București (~35%), Brașov (~12%), Cluj-Napoca (~10%) and Timișoara (~8%). Without exception, every one of them operates on the same model: either the customer uploads an STL or 3MF they designed themselves or found on Thingiverse, or they pay a design fee for a specialist who works from imprecise text descriptions.

The result is that most people who would benefit from a custom part never end up ordering one. The Romanian desktop 3D printing market is projected to reach 48.8M USD by 2030 at a 22% annual growth rate, and this blocked segment is the clearest growth opportunity with no direct competition.

The challenge

We started from a single hypothesis: if we can remove the STL file from the ordering flow, we open the service to a customer segment ten times larger. To do that, we needed three components that no competitor had assembled together:

• a measurement method that doesn't require the customer to own a specialised instrument,
• a library of parametric templates flexible enough to cover most household and office needs,
• a production infrastructure that fulfils the order at tight tolerances and inside a price range an individual buyer will accept.

Measurement from a photo

The core of the platform is the /masoara flow. The customer takes 1-5 photos of the area where the part will sit, with a standard credit card in the frame (85.60 × 53.98 mm, fixed dimensions per ISO/IEC 7810 ID-1). The vision pipeline uses the card as a scale reference, applies perspective rectification and extracts the relevant distances at a target precision of ±2 mm.

The 2 mm tolerance isn't arbitrary. It's the threshold where a shelf bracket fits without forcing, where a door stopper doesn't scrape, and where a replacement knob seats correctly on the original shaft. Below this tolerance, photo measurement drifts against the real variation of the physical space, so we chose to expose it transparently in the UI: every extracted value shows a confidence score, and the customer can manually correct any dimension before ordering.

The pipeline details, calibration criteria and fallback steps for hard cases are documented on the Photo measurement guide.

Parametric templates, not STLs

We built the initial library from parts with real demand, validated against Ahrefs search data for the RO market: L-brackets, wall hooks, cable clips, cylindrical and rectangular spacers, round replacement knobs, drawer dividers, desk organisers and enclosure boxes. Each belongs to a functional category: replacement, home, office, auto, workshop and display.

Each template exposes a bounded set of parameters: width, height, thickness, inner diameter, angle, colour. The configurator applies real-time validations (minimum and maximum bounds, geometric collisions) and shows the final price before payment, with no hidden costs. The parametric model has a useful SEO side effect: the product page is a real content page, not a filter result, and each template becomes independently indexable for specific commercial queries.

Production: Bambu Lab and real materials

Production runs on the Bambu Lab P2S, with two calibrated materials at launch: PLA for decorative or lightly loaded parts, and PETG for parts with mechanical requirements or moisture exposure. Material choice is guided by a dedicated resource, PLA vs PETG, linked from the configurator at the moment it matters.

We picked Bambu Lab for tolerance repeatability, print speed on small parts and self-calibrating mechanics, which reduces human defect risk at low volumes. The reference price of a part is described openly in the how much does a 3D printed part cost guide, with the fixed components (setup, filament, shipping) made visible.

The complete order flow

1. Photograph. The customer opens /masoara, follows the overlay guidance and uploads 1-5 photos with the credit card in the frame.
2. Customise. The system proposes the closest template from the catalog, with parameters pre-filled. The customer can correct any dimension and sees the price update in real time.
3. Order. Payment happens online, and the final dimensions flow directly into the production job, with no manual step in between.
4. Receive. The part is printed, visually inspected and shipped via courier. Target window: 3-5 business days from order.

Technical architecture in brief

The frontend is Angular with SSR, which gives a full HTML response on the first request for every indexable page (product, category, guide). The backend is Spring Boot on PostgreSQL. The vision pipeline runs in a dedicated module, called synchronously for measurement previews and asynchronously for cases that need review. The product configurator is the only conversion component tolerated on commercial routes, precisely because it is both the competitive moat and the component that carries ranking for "order a part at dimension" queries.

The UI is bilingual RO/EN from day one, and the routing taxonomy follows the same convention we use on Elisara: Romanian routes for the domestic market, English routes for the developer flow and international visitors. Every product, category and guide has its own canonical with bidirectional hreflang, so the RO and EN versions don't cannibalise each other in SERP.

Technical SEO decisions that kept us off the ceiling

We followed three strict rules that we apply across the portfolio. First: mandatory SSR for every public page, with full HTML on the first request. Without SSR, otherwise well-structured pages get stuck in "discovered, currently not indexed" for weeks.

Second: a minimum content threshold for indexing. Categories with fewer than three products and guide pages under 800 characters are noindexed until they clear the bar, so we don't dilute domain authority with thin pages. Catalog filters (query string) canonicalise to the unfiltered variant and stay noindex regardless of the filtered content, so they don't compete with the real category pages.

Third: structured JSON-LD on every critical route. Products carry schema.org Product with Offer and BreadcrumbList, guides carry Article plus FAQPage, the homepage carries Organization. The components that emit structured data are centralised in a single service with cleanup in ngOnDestroy, so client-side navigation doesn't leave residual JSON-LD in the DOM.

What we learned

The most valuable lesson was that the template library has to be built around search demand, not around what felt intuitive at the start. We validated the initial list against Ahrefs data for the RO market: car phone holder (9,600 monthly searches), door stopper (5,300), furniture handles (4,100), furniture legs (3,400), laptop stand (2,400). All at KD zero, meaning well-built product pages can rank without real competition. Without this validation, the risk was building templates that were technically impressive but tied to search volume nobody was actually generating.

The result

Tipografia3D is live. It's the first Romanian 3D printing service that never asks you for an STL file, and operationally it combines pieces rarely found together: a catalog with a final price, a full parametric configurator, photo-based measurement and fast delivery. If you have a part in mind but no file for it, you can start directly from the measurement flow.