Thermal Twin

The description states that Thermal Twin builds a physics-based digital twin of buildings using:

• A 2D floor plan
• Existing data (indoor/outdoor temperature, heating)
• No additional sensors

It claims to:

• Identify a physics-constrained thermal twin
• Detect when the building drifts from its calibrated behavior
• Simulate renovation scenarios with cost, CO₂, payback, and subsidies
• Provide a shared source of truth for four trades (owner, engineer, architect, operator)
• Allow interrogation via natural language through GPT-5.6

The system is built using:

• FastAPI backend
• Vanilla JavaScript UI with MapLibre and deck.gl for 3D rendering
• A resistance–capacitance (RC) grey-box network model
• Matrix exponential discretization
• L-BFGS-B optimization in log-space
• Bayesian Information Criterion (BIC) for structure selection

It integrates with:

• BDNB, ADEME DPE, IGN BD TOPO databases
• PLEIAData dataset
• BDNB/DPE records of a real French social-housing building

Not evidenced No mention of actual deployment, user feedback, or performance metrics beyond the development process.

The description states that Thermal Twin was developed to address:

• High cost and time of traditional energy audits
• Need for site visits and sensor installation
• Lack of shared information across trades in building renovation

It positions itself as a solution that:

• Eliminates need for sensors
• Reduces audit time from months to minutes
• Enables real-time collaboration among stakeholders
• Uses AI to make diagnostics accessible to non-experts

The author claims the tool is based on a peer-reviewed method selected for a Best Paper Award at SASBE / BDT 2026, and that it was turned into a working prototype in days using Codex.

Inference This suggests a shift from academic research to productization — but no evidence of how the tool is positioned in the market or whether it has been tested beyond the hackathon.

The description states:

• The tool is built for the French building sector
• It uses public French databases (BDNB, ADEME DPE, IGN BD TOPO)
• It targets owners, engineers, architects, and operators of buildings
• It focuses on renovation diagnostics, particularly in social housing

It also mentions:

• A pilot with a French social landlord
• Plans to plug into BIM/BMS systems
• Contextualization for region-specific subsidies

Not evidenced No information about actual customers, customer segments, or whether the tool is being used by any of the target actors.

The description does not state:

• Any pricing model
• Revenue streams
• Monetization strategy
• Subscription or licensing details

It mentions:

• The tool is built for French public databases
• It simulates renovation scenarios with cost, CO₂, payback, and subsidies
• It integrates with BDNB/DPE records

Inference It may be positioned as a diagnostic tool for building owners or public agencies, but no business model is described.

The system uses:

• FastAPI backend
• Vanilla JavaScript UI
• MapLibre + deck.gl for 3D rendering
• GPT-5.6 for grounding and QA
• Codex to reverse-engineer the method from a notebook
• A physics-based RC grey-box model with matrix exponential discretization
• L-BFGS-B optimization and BIC structure selection

It claims:

• The twin is identified on the PLEIAData dataset
• It uses real geometry from IGN BD TOPO WFS
• It has a 745 MB dataset shipped via Release
• It includes 87 passing tests

Not evidenced No information about scalability, performance, or production deployment.

The description states:

• The method was selected for a Best Paper Award at SASBE / BDT 2026
• It was built in days using Codex
• It includes 87 passing tests
• It has a GitHub repo with peer-reviewed paper and demo
• It uses real data from BDNB, ADEME DPE, IGN BD TOPO

It also mentions:

• A pilot with a French social landlord
• Plans for semantic interoperability, portfolio-scale views, and BIM/BMS integration

Not evidenced No revenue, customers, or adoption beyond the hackathon prototype. No evidence of real-world usage or impact.

The description does not mention:

• Direct competitors
• Market size or landscape
• Existing tools in building digital twins or energy audits

It implies that current solutions require:

• Site visits
• Sensors
• Manual document handoffs
• High cost and time

Inference Thermal Twin positions itself as an alternative to traditional, sensor-based, and document-heavy audit methods. However, no competitive analysis is provided.

• No traction or revenue: The tool is described only as a hackathon prototype.
• Unverified claims: The description states that the system is based on peer-reviewed research but does not confirm whether it has been validated in practice.
• AI grounding risks: While GPT-5.6 is used for grounding, there’s no evidence of how well it avoids hallucinations or maintains accuracy.
• Uncertainty handling: The tool surfaces uncertainty rather than confident numbers — this may be a product principle but could also be a limitation for users seeking certainty.
• Limited scope: It is built for the French building sector and uses only French public data sources.

1. What is the actual validation process for the digital twin? Has it been tested on real buildings beyond the demo?
2. How does the tool handle uncertainty in its models — is this a feature or a limitation?
3. Are there any partnerships with French public agencies or landlords already in place?
4. What are the technical limitations of scaling this to larger portfolios or different countries?
5. How is the GPT-5.6 integration grounded — what safeguards prevent hallucinations?
6. Is there an internal team or external validation of the model’s accuracy?
7. What is the long-term vision for monetization and product roadmap?