skima-Architectural CAD for AI agents

The description states that Skima is an AI-native architectural CAD engine. It integrates GPT-5.6 with a structured design workflow involving:

• Site data collection from Korean addresses (geospatial, parcel boundaries, land-use info)
• Regulatory reference criteria translated into computational rules and geometric constraints
• Generation of editable CAD drawings using validated proposals
• Validation of geometry, area, overlaps, access relationships, and building coverage ratio
• Revision handling through PDF markups interpreted by GPT-5.6

It uses technologies such as canvaskit, three.js, react, express.js, postgresql, and openai (specifically GPT-5.6), among others.

Confidence Low — this is a self-reported technical architecture without evidence of deployment or usage.

The description positions Skima as an AI-native CAD engine that bridges the gap between generative AI outputs and professional architectural practice by enabling editable, rule-informed CAD drawings. It claims to:

• Translate design intent into editable CAD objects
• Visualize regulations spatially (e.g., building coverage ratio, daylight setbacks)
• Produce professional architectural drawings with layers, line weights, hatches, dimensions, etc.
• Turn PDF markups into validated CAD revisions

The author also states that Skima is built for use in South Korea and integrates local zoning laws and ordinances.

Confidence Low — all claims are self-reported and lack independent verification or demonstration of traction.

The description implies the primary user is an architect working on site planning in South Korea. The system is designed to:

• Understand architectural design intent
• Work within real site conditions and building codes
• Generate and revise editable CAD drawings

There is no mention of other stakeholders such as engineers, developers, or clients.

Confidence Low — no evidence of customer personas, user interviews, or market segmentation beyond the author’s own account.

The description does not provide any information about pricing models, monetization strategies, or business model assumptions. It also lacks details on whether Skima is intended for individual use, firm licensing, or marketplace integration.

Confidence Not evidenced — no commercial or financial data provided.

Skima uses a multi-toolchain approach involving:

• GPT-5.6 as the core AI agent
• MCP-based architectural design toolchain
• Codex for structuring laws and ordinances into computable rules
• CanvasKit, Three.js, React, PostgreSQL for rendering and storage
• PDF viewer with markup interpretation capabilities

It implements a generate–validate–repair–approve loop to ensure CAD consistency and compliance.

Confidence Medium — the technical stack is described in detail but not independently verified or tested in production.

The description states that Skima was built for the OpenAI 2026 hackathon, suggesting it is a prototype or proof-of-concept. There is no evidence of:

• Revenue
• Customers
• Product adoption
• Market traction
• Iteration history or product maturity beyond its initial development phase

Confidence Not evidenced — this is a self-reported project with no external validation.

The description does not mention any competitors or existing solutions in the architectural AI/CAD space. It does not describe how Skima compares to other tools or platforms used by architects for design, planning, or code compliance.

Confidence Not evidenced — no competitive analysis or market positioning provided.

• Unverified claims: All features and capabilities are self-reported without external validation.
• Limited scope: The system is built for South Korea only; no indication of scalability or international expansion plans.
• Prototype nature: Built for a hackathon, not yet proven in real-world use.
• Dependency on AI hallucination risk: Reliance on GPT-5.6 for design decisions introduces potential inaccuracies unless tightly controlled.
• No commercialization path: No mention of monetization, partnerships, or go-to-market strategy.

Confidence High — these are logical inferences from the lack of evidence and self-reporting nature of the description.

1. What is the current stage of development? Is this a working prototype or a pre-alpha version?
2. Are there any early users or pilot partners in South Korea?
3. How does Skima handle jurisdictional differences beyond South Korea?
4. What are the specific validation rules used for different types of building codes and zoning laws?
5. Has the team tested the system with actual architects or firms?
6. Is there a plan to expand beyond site planning into full architectural design workflows?
7. What is the long-term vision for monetization or product commercialization?