Pipeline and datasets for data-centric AI on real-world floor plans
Details
- External ID
- 46896474
- Source
- HN
- Company
- —
- Product
- Pipeline and datasets for data-centric AI on real-world floor plans
- Website domain
- standfest.science
- Launched
- Feb. 5, 2026
- Cohort
- —
- Upvotes
- 12
- Upvotes percentile
- 0.5970350404312669
- Tags
- —
- Fetched at
- Sept. 7, 2026, 9:26 p.m.
- Updated at
- Sept. 7, 2026, 9:26 p.m.
Enrichment
- Theme
- embodied AI and robotics platforms
- Vertical
- Construction
- Function
- Data infrastructure
- Audience
- Developer
- AI stance
- AI feature
- Project type
- Commercial product
- Normalized one-liner
- training data pipeline for floor plan analysis
- Manually corrected
- False
Could you build this?
No Creating a real-world architectural dataset and computer vision data pipeline for complex floor plans requires specialized CAD/BIM domain knowledge, licensed architectural data access, and bespoke geometric processing algorithms.
What it would actually take: The system requires vectorization pipelines capable of converting raster architectural blueprints and complex IFC/CAD/BIM files into normalized geometric graphs, room connectivity graphs, and spatial polygon datasets. The core bottleneck is obtaining tens of thousands of proprietary real-world building floor plans, navigating Swiss zoning/property data privacy restrictions, and engineering robust heuristic/geometric graph-extraction algorithms to reconcile noisy vector drawings. It requires significant domain expertise in architectural geometry, CAD standards, and spatial computer vision.
Discussion
4 comments analyzed.
Competitors mentioned: U-Net models for floor plan segmentation, Transformer-based approaches for layout detection, Computer vision (CV) early approaches
Concerns raised: Floor plan generators don't preserve critical constraints (scale, closure, topology, entrances), Outputs look plausible but fail when used as geometry for downstream tasks, Real plan archives are adversarial input (old drawings, multiple styles, low-res scans, conflicting versions), Small errors in geometry compound rapidly for downstream tasks like energy modeling or digital twins, Symbol ambiguity (stairs vs ramps, doors vs windows) requires manual resolution
Feature requests: Context graph integration (spatial topology, semantics, building/unit/site constraints, environmental context), Cross-floor consistency enforcement, Automated EDA on datasets (distributions, correlations, biases, failure modes mapping), Cross-domain generalization between architectural styles
Competitors
Other products that read as similar to this one — 1769 launches clear the similarity bar, closest 8 shown.
Attention rank: #612 of 1770 (itself plus its competitors, highest first — normalized so YC and Product Hunt are compared fairly).
Launched 99 days after the earliest competitor.
- 🦾 Cortex AI — Large-Scale Real-World Data for General-Purpose Robotics · yc · 2025-11-08 · 20 upvotes · similarity 0.62
- Avoice: The AI workspace for architects · yc · 2026-02-11 · 71 upvotes · similarity 0.60
- Spatial AI: Large datasets for robot foundation models · yc · 2025-11-17 · 22 upvotes · similarity 0.59
- Avoice: Harvey for Architecture · yc · 2026-03-08 · 38 upvotes · similarity 0.59
- Hebbian Robotics - APIs for searching and analyzing Physical AI data at scale · yc · 2026-08-09 · 81 upvotes · similarity 0.57
- Vision Lab: The Industrial Data Layer for Robotics · yc · 2026-03-23 · 18 upvotes · similarity 0.57
- Feldy: Field-first AI for home services and contractors. · yc · 2026-08-13 · 22 upvotes · similarity 0.57
- DecisionBox · hn · 2026-04-23 · 6 upvotes · similarity 0.57
Other launches for this product
- No other launches for this product.
Same idea, different domain
Nobody's really built a data infrastructure tool for Media & entertainment yet.