Valentina Mayer-Steudte
Co-Founder, CEO
Company leadership, strategy & product management
M.Sc. Architecture: strategic planning, BIM, business development, branding.
Experience: Co-Founder Wertstudent, Bez+Kock Architekten, cruu architecture (BIM)
- BoQ
- Position
- Teilleistung
- Titel
- Mengenansatz
- IFC / BIM
- IfcWall
- IfcSlab
- IfcBeam
- IfcSpace
- Ökobaudat
- GWP
- EPD
- Lebenszyklus
- Datensatz
- Cost data
- Kostenkennwert
- Kostengruppe
- Standardposition
- KG 300
- Rohbau
- Bodenplatte
- Stütze
- Geschossdecke
- Fundament
- Brandschutz
- Brandabschnitt
- Entrauchung
- F90
- Sprinkler
- Bewehrung
- Betonstahl
- Bewehrungsmatte
- Bügel
- Stabstahl
- Trockenbau
- Ständerwand
- Gipskarton
- Abhangdecke
- Vorsatzschale
- TGA
- Lüftung
- Heizung
- Sanitär
- Elektro
- Estrich
- Zementestrich
- Anhydritestrich
- Dämmschicht
- Randstreifen
- Mauerwerk
- Kalksandstein
- Ziegel
- Porenbeton
- Mörtel
- Dämmung
- Mineralwolle
- EPS
- Perimeterdämmung
- WLG 035
Domain-Specific AI Infrastructure · API
Semantic infrastructure for construction data.
One API for construction-specific OCR, embeddings and reranking. Make documents machine-readable, connect data across formats and rank results by relevance in search, matching and RAG. Built and hosted in Europe.
curl https://api.flinq.ai/v1/embeddings \
-H "Authorization: Bearer $FLINQ_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "flinq-embed",
"input": "Stahlbeton C25/30, Bodenplatte d=25 cm"
}'Conceptual example. Runnable examples live in the docs.
Available as a cloud API hosted in the EU (GDPR-compliant)
+30%
better product retrieval than OpenAI
nDCG 0.78 vs. 0.60
ρ 0.71 #1
understands units of measure better than the whole field
vs. ρ 0.57 (best model in the field)
83%
recognises components by their 3D shape
82.9% accuracy; text-only models: 6.3%
How it works
One API for all construction data
flinq makes construction documents machine-readable, links data by meaning and surfaces the right information for your applications.
-
Step 01
Construction documents and data as input
Tender texts, bills of quantities, BIM models, plans, photos and cost data, as they are.
Data formats -
Step 02
Making documents machine-readable
OCR turns scans and images into text and structure. Data that is already structured skips this step.
OCR -
Step 03
Semantically linking data and finding the right information
The embedding links content semantically in one shared vector space, the reranker sorts the results by relevance.
Embedding + Reranker -
Step 04
You build on top
In your own software, through one API.
Use cases
What it gives you
Why software providers build on flinq
-
01
Construction-specific
Trained for construction terminology, units, standards, cross references, and document layouts. Improves extraction, matching, retrieval, and relevance ranking across construction data.
-
02
Efficient
Task-specific models handle extraction, retrieval, and ranking without running a general-purpose LLM for every request. This reduces development effort, token usage, and compute costs.
-
03
Modular
OCR, embedding, and reranking capabilities are available through individual APIs. Software providers integrate only what they need into existing products and workflows.
-
04
European
Trained on European construction data and hosted in the EU. Transparent data processing supports data sovereignty and compliance evidence.
How it compares
Why a domain-specific retrieval infrastructure?
Generic models read construction documents as arbitrary text. flinq is trained on the data, formats and terminology of the construction industry, and the difference shows up where it matters: in what comes back, and whether you can rely on it.
Measurably ahead of general-purpose models
On the BauSatz benchmark flinq reaches nDCG 0.78 on product retrieval against 0.60 for OpenAI. Fewer false matches mean less manual rework.
See the benchmarkReproducible results you can defend
Same input, same vector, every time. Traceable for costing and tenders, where generative models answer differently on every call.
Dimensions, standards and technical differences are preserved.
Trained on construction data: dimensions, grades, units and standards references help decide what counts as a hit.
One integration, several features
The same vectors power search, matching, RAG, deduplication and clustering. You integrate once and cover several features with it.
Fits into the processes you already run.
flinq is an API, not a platform. The functions run inside the software your team already uses, and the results show up in the existing workflow. No new tool to roll out, no data migration, no relearning.
Processing and hosting remain within the EU.
GDPR-compliant in the EU, without a detour via a US provider.
nDCG figures from the BauSatz benchmark (zero-shot); the full table is further down this page.
Benchmark
BauSatz: flinq small vs. the field
flinq small is a single, compact model. On the BauSatz benchmark it leads on six of eight tasks against larger, zero-shot models.
| Task | flinq small | OpenAI | Qwen3-8B | Mistral | Gemma2 | BGE-M3 |
|---|---|---|---|---|---|---|
| Product retrieval | 78.4 | 59.5 | 56.8 | 40.3 | 58.6 | 36.8 |
| Unit semantics | 70.9 | 55.8 | 58.7 | 52.5 | 57.8 | 48.6 |
| Spec reranking | 78.5 | 72.6 | 64.9 | 70.7 | 61.8 | 68.9 |
| IFC name matching | 71.2 | 13.3 | 22.2 | 19.1 | 13.5 | 49.7 |
| Trade clustering | 36.1 | 33.1 | 31.3 | 29.1 | 30.5 | 24.3 |
| Equivalence pairs | 97.1 | 96.1 | 96.1 | 96.1 | 96.1 | 96.1 |
| Section classification | 72.8 | 71.5 | 73.2 | 60.7 | 72.4 | 69.5 |
| Position matching | 85.2 | 87.1 | 85.7 | 87.2 | 86.1 | 86.9 |
For developers
Access via an HTTP API
Access to the model is via an HTTP API. A request embeds one or more construction objects, and the response returns the corresponding vectors.
{
"model": "flinq-embed",
"input": [
"Stahlbeton C25/30, Bodenplatte d=25 cm",
"Bewehrungsstahl BSt 500 S, Mattenbewehrung"
]
}A request embeds a bill-of-quantities line item.
{
"model": "flinq-embed",
"data": [
{ "index": 0, "embedding": [0.0131, -0.0427, 0.0588, …] },
{ "index": 1, "embedding": [0.0204, -0.0119, 0.0731, …] }
]
}The response returns the vector and the model id. Vector truncated.
These are conceptual examples that show the format. The full reference and runnable examples live in the docs.
Go to the API reference Deployment and trust
As a cloud API, hosted in the EU
flinq is available as a cloud API. Operation is GDPR-compliant and hosting is in the EU.
Cloud API
Direct access via the hosted API. No infrastructure to run yourself.
GDPR-compliant
Data processing follows the requirements of the GDPR.
EU hosting
The hosted API runs in the EU.
Team
The team behind flinq
flinq is built by a team with backgrounds in architecture, BIM, machine learning and construction economics.
Matthias Hornung
Co-Founder, CTO
Machine learning, technology & software development
M.Sc. ITECH and M.Sc. Architecture: BIM, graph ML, robotics & software.
Experience: Design-to-Production, Transsolar, Blocher und Partners/Schwarz Architekten
Lukas Simon
Co-Founder, CPO
Product development, data pipelines & data analysis
cand. M.Sc. Economics: data analysis, financial modeling, process optimization.
Experience: Mercedes Benz, ZEW
Supported by partners and programmes from the ecosystem.
News & updates
Latest developments
Stay up to date with new features, insights and best practices.
flinq receives the EXIST start-up grant
Since August, the German Federal Ministry for Economic Affairs and Energy and the European Union have been funding a full year of development. The team continues to build the cross-modal embedding model for construction, supported by the Institute of Construction Economics at the University of Stuttgart and Prof. Christian Stoy.
Read on LinkedInSoftware is getting easier to build. Domain-specific intelligence is not.
BIM models, bills of quantities, product data and technical documents describe the same physical reality, yet follow different structures, standards and vocabularies. flinq maps them into one shared semantic space, trained on European infrastructure with HammerHAI at HLRS Stuttgart. The text-based beta is available today; the first cross-modal version follows in September.
Read on LinkedInflinq at the ZIA PropTech breakfast at S-TEC
Invited by the ZIA at S-TEC and Fraunhofer IAO: the conversations confirmed that companies are not looking for yet another off-the-shelf tool, but technology that understands their workflows and fits in. An announcement on this follows in the coming weeks.
Read on LinkedIn FAQ
Frequently asked questions
What is flinq?
flinq is the embedding API for the European construction industry. A cross-modal model maps heterogeneous construction data into one shared vector space, from bills of quantities and BIM models to site photos, manufacturer products, cost data and standards. These document types exist across Europe in different national formats; today flinq supports the German formats such as GAEB most deeply. Access is via an API.
Is flinq a SaaS product or an API?
flinq is an API service. The product is the model behind an API, not a finished application. Bill-of-quantities matching, cost prediction and RAG are examples of what you build on the API, not features flinq ships.
Which data and file formats does flinq support?
flinq reads common construction file formats and the domain content they carry. These are GAEB, IFC, PDF, JPG, PNG and CSV. From them the model recognises, among others, bills of quantities and tender texts, BIM models and building elements (including 3D geometry), site photos, manufacturer products, cost data and standards. GAEB here is the German format for bills of quantities, one national instance of a document type found across Europe, whereas IFC is an international BIM standard. Support is deepest today for the German formats. All are embedded into the same vector space.
How do I access the model?
Via an HTTP API. You send construction objects as a request and receive embedding vectors in return. The full reference and runnable examples live in the docs.
What can I build on flinq?
Typical examples are matching bill-of-quantities line items, early-phase cost prediction and RAG over construction data. Building on that, you can identify VOB-compliant tender texts or price bills of quantities via external cost databases. Further examples are pricing tendered line items via an internal cost database, matching manufacturer products to line items, classifying products by ETIM, and retrieving the relevant national standards, such as DIN. You build these applications on the API; flinq provides the embedding model, not the finished application.
What sets flinq apart from generic embedding models?
flinq is trained on domain-specific tasks. On the BauSatz benchmark, flinq small reaches nDCG 0.78 on product retrieval versus 0.60 for OpenAI, and recognises components by their 3D shape with 82.9% accuracy, while text-only models reach 6.3%.
Is flinq GDPR-compliant?
Yes. flinq is GDPR-compliant and hosted in the EU.
What is the quality of the embeddings based on?
On a curated, cross-modal dataset of matched construction objects. This dataset is the defensible core. The model backbone is swappable.
Question not answered here?
Contact us