Agentic AI-Enabled Framework for Thermal Comfort and Building Energy Assessment in Tropical Urban Neighborhoods

Abstract: In response to the urban heat island effects and building energy demands in Singapore, this study proposes an agentic AI-enabled reasoning framework that integrates LLMs with lightweight physics-based models. Through prompt customization, the LLMs interpret urban design tasks, extract relevant policies, and activate appropriate physics-based models for evaluation, forming a closed-loop reasoning-action process. These lightweight physics-based models leverage core thermal and airflow principles, streamlining conventional models to reduce computational time while predicting microclimate variables, such as building surface temperature, ground radiant heat, and airflow conditions, thereby enabling the estimation of thermal comfort indices, e.g., physiological equivalent temperature (PET), and building energy usage. This framework allows users to explore a variety of climate-resilient building surface strategies, e.g., green façades and cool paint applications, that improve thermal comfort while reducing wall heat gain and energy demand. By combining the autonomous reasoning capacity of LLMs with the rapid quantitative evaluation of lightweight physics-based models, the proposed system demonstrates potential for cross-disciplinary applications in sustainable urban design, indoor-outdoor environmental integration, and climate adaptation planning. The source code and data used in this study are available at: https://github.com/PgUpDn/urban-cooling-agent.

• The paper showcases an AI-enabled framework that couples LLM reasoning with physics-based simulations to autonomously assess thermal comfort and energy performance in tropical urban neighborhoods.
• It employs a modular, LangGraph-based state machine to execute stages like geometry validation, solver orchestration, and material strategy recommendation for design audits.
• Key findings reveal an ‘albedo penalty’ trade-off where energy savings might increase pedestrian thermal discomfort, guiding targeted urban interventions.

Agentic AI-Enabled Reasoning for Thermal Comfort and Energy Evaluation in Tropical Urban Contexts

Framework Architecture and Integration

The paper presents a modular agentic AI-enabled reasoning framework that couples LLMs with lightweight physics-based simulation models to streamline assessment of thermal comfort and building energy in dense tropical urban neighborhoods (2604.21787). Central to the architecture is a LangGraph-based state machine that tightly orchestrates five stages: intent analysis, geometry validation, solver orchestration, material strategy recommendation, and integrated reporting. The LLM is advisory, parsing natural language queries for urban design tasks, dynamically extracting relevant regulatory or climatic policies, and determining appropriate simulation workflows. Parameter governance ensures authoritative boundary conditions by strict priority—configuration defaults, climate databases (IWEC), real-time meteorological APIs, context-aware LLM suggestions, and final user overrides—captured in auditable JSON logs.

The geometric input is standardized via STL meshes, with automated cleaning, centroid-based ID assignment, and spatial grid discretization (average $\sim$2 m). Solver backends are abstracted, supporting rapid switching from lightweight models to high-fidelity simulators. Every LLM transaction and parameter merge is logged for traceability, enabling reproducibility and reviewability across cross-disciplinary teams.

Physics-Based Modeling Components

Three key simulation modules underpin quantitative assessment. The CFD solver implements a 2D potential flow model discretized across height slices, interpolating wind vectors on complex 3D building envelopes, and adjusting local microclimate (temperature, humidity) via empirical tropical boundary layer models. Radiative transfer is computed via transient energy balance: coupling shortwave and longwave fluxes, convective heat transfer, and thermal storage, yielding accurate surface temperature and pedestrian-level mean radiant temperature (MRT). Thermal comfort is quantified using PET derived from MEMI, directly integrating CFD and radiation outputs with physiological parameters. Cooling energy is measured via a 1D CTF model, aggregating exterior heat flux for each mesh element and normalizing Energy Use Intensity (EUI), maintaining sufficient comparative fidelity for design trade-off analysis and hotspot identification.

Autonomous Diagnostic and Mitigation Workflows

The framework is validated in high-density urban scenarios in Singapore across two distinct workflows: (1) Baseline diagnostic audit and (2) Closed-loop material intervention.

Scenario I (Baseline Audit): The Orchestrator Agent interprets the "inter-monsoon" query, autonomously selects representative dates, configures simulation domains, and identifies critical comfort and energy hotspots. In both districts analyzed, PET peaks exceeded 52°C at midday in sun-exposed plazas, confirmed by spatial thermal maps. High-rise structures were flagged as cooling demand outliers, normalized by envelope area. The system accurately traced causal chains—combining sky view factor, stagnated airflow, and façade orientation—linking each hotspot to actionable interventions (shading, façade optimization).

Scenario II (Material Intervention): The Orchestrator Agent applies high-albedo coatings per ASHRAE standards to roofs/walls/ground, performing a baseline-to-mitigation delta analysis. Key findings include:

• Building energy reduction: Envelope cooling loads dropped significantly (peak reductions: up to 11.4% in daily energy in prioritized towers).
• "Albedo penalty": Despite reduced surface temperatures and energy gains, PET at noon increased by ~1°C due to elevated reflected shortwave radiation at pedestrian level, revealing a crucial trade-off between envelope cooling and outdoor comfort.

The Recommendation Agent synthesizes this feedback, advocating high-albedo roof coatings for energy mitigation, but cautioning against high-reflectivity pavements, thereby decoupling strategies for indoor/outdoor optimization.

Key Numerical Outcomes and Contradictory Findings

• Thermal comfort: PET hotspots are consistently identified, with severity >50°C in sun-exposed plazas.
• Energy demand: Tall towers show 8.5%–11.4% reduction in daily cooling energy with material interventions.
• Comfort-energy trade-off: In several cases, increased ground albedo improves building cooling yet raises PET for pedestrians, directly contradicting intuitive design assumptions. The system successfully navigates and flags these nonlinear trade-offs during autonomous reasoning cycles.

Practical and Theoretical Implications

The agentic workflow demonstrated in the paper advances urban microclimate and energy analysis by reducing specialist dependence and accelerating cross-scenario audits while maintaining deterministic workflow traceability. Integrating LLM reasoning "on top" of physics solvers without overwriting domain logic enables conversational setup, rapid diagnostics, and parameter provenance. Backend-agnostic orchestration enables flexible scaling—from rapid exploratory analyses to resource-intensive validation—without architectural changes.

From a theoretical perspective, the discovery of the "albedo penalty" underscores that microclimate and energy interventions in tropical cities require explicit trade-off modeling, rather than heuristic optimization. The framework’s autonomous reasoning loop can uncover emergent physical effects, supporting policy, planning, and design teams with actionable, evidence-based recommendations tied to simulation outputs.

Outlook and Future Directions

Future enhancements are specified, including scalable meshing and tiling for large-area simulation, automated computation of regulatory KPIs, and the development of interactive dashboards for live auditability and user engagement. Standardizing comfort and energy metrics (Lawson/NEN threshold checks) will facilitate direct regulatory compliance integration. Real-time reporting and parameter traceability can democratize expert analysis, enabling non-specialists to interrogate simulation provenance.

The modular, backend-agnostic structure supports extension toward full-fidelity solvers and AI surrogate models, expanding applicability for integrated climate adaptation, indoor-outdoor environmental design, and urban resilience planning in tropical contexts.

Conclusion

This agentic AI-enabled framework sets forth an auditable, deterministic workflow for thermal comfort and building energy evaluation in tropical urban neighborhoods by integrating autonomous LLM reasoning with lightweight physics-based simulation. Critical results demonstrate both the efficacy and pitfalls of standard mitigation strategies, revealing non-intuitive physical artifacts such as the "albedo penalty." The architecture accommodates rapid scenario audits and closed-loop intervention cycles, promising scalable, transparent, and reproducible application in sustainable urban design and climate-resilient planning.