Accurate State of Charge (SOC) estimation is a critical requirement for the safe and efficient operation of lithium iron phosphate (LFP) battery systems. Due to the flat open-circuit voltage (OCV) profile of LFP chemistry, voltage-based estimation alone is insufficient, particularly under dynamic current conditions. Purely data-driven methods offer high accuracy but lack physical interpretability, while model-based approaches ensure consistency with electrochemical behavior but remain sensitive to modelling inaccuracies. This thesis develops, implements, and experimentally validates a complete hybrid SOC estimation framework that integrates an Equivalent Circuit Model (R0+1RC) with an Extended Kalman Filter (EKF) and a neural-network-based correction layer. The workflow follows a structured and modular pipeline: dataset preprocessing and reference SOC reconstruction; OCV characterization; ECM parameterization; EKF implementation; and finally, supervised learning of residual SOC errors using a recurrent neural network (GRU-based architecture).