Buildings account for 40% of global energy consumption, with heating, ventilation, and air conditioning (HVAC) system contributing significantly [1]. A precise prediction of the thermal load helps reduce energy waste in HVAC systems by means of improved techniques. Especially in office buildings, occupancy conditions have a significant impact on the heating or cooling load of a building [2]. Therefore, it is necessary to integrate real-time estimation of occupancy conditions and ventilation into a thermal digital twin of an office building to achieve energy saving goals. This study develops a coupled temperature–CO2 network for multi-zone buildings that explicitly accounts for airflow-driven convective heat transport in addition to conductive heat transfer. The building is represented as a lumped (RC) thermal network coupled with a CO2 transport network. CO2 measurements are used as a non-intrusive proxy to infer latent occupancy and ventilation/airflow conditions, which then drive both gas transport and convective heat exchange. A single-floor layout is constructed with four office rooms, two larger rooms, and a connecting hallway, with integrated CO2 and temperature sensors. Both synthetic and scaled physical validation experiments were implemented based on the layout. Model parameters and time-varying operating conditions are estimated using a moving-window Bayesian inference framework, enabling uncertainty-aware reconstruction and short-term prediction of zonal temperature and CO2 trajectories. Results show that incorporating convective coupling improves temperature prediction during airflow regime changes and intervention phases, while the inferred airflow and source terms remain physically interpretable and track operational switching. The proposed framework supports robust calibration for monitoring and energy-performance assessment under limited sensing and uncertain occupancy.