Pyrolysis of biomass is a challenging technique to convert raw biomass to liquid fuels. Fluidized beds are often implemented for this application due to their high heat and mass transfer capabilities. The process is often considered as a multiscale problem. Biomass composition impacts the production rate of gases at a molecular scale. The size, shape and porosity of the biomass particle influence the heat and mass transport at a particle scale. Hydrodynamics at reactor scale determines the particle mixing, and heating rates inside a reactor. A multiscale modeling strategy is necessary to accurately model the Fluidized Bed Reactors (FBRs). Improving the quality of biooil necessitates catalytic upgradation of the vapours coming out of fluidized beds. The operation is carried out in gas-solid catalytic risers. Mesoscale structures such as particle clusters in the risers can impact heat and mass transfer rate, and reduce conversion. Developing a computationally fast technique to capture and quantify the mesoscale structure is crucial in gas-solid multiphase systems. Moreover, the CFD simulations of catalytic risers are extremely time consuming. This requires building computationally fast surrogate models to predict conversions based on the mesoscale information and reaction kinetics. This work comprises of developing a first principle based multiscale modeling strategy to simulate biomass pyrolysis in fluidized beds, implementation of a computationally fast opensource Machine Learning (ML) technique to capture mesoscale structures, and building a surrogate model based on first principle CFD simulations of gas-solid reactive systems.