Accurate State of Health (SOH) estimation is essential for optimizing lithium-ion battery performance, especially in second-life applications. This paper introduces a fast SOH estimation method using the first 10 seconds of charge voltage data, followed by polynomial regression to extract features for AI algorithms. A hybrid model combining Feed-Forward Neural Networks (FNN) and Gaussian Process Regression (GPR) is proposed. This model was more accurate in detecting the SOH using just 10 seconds of charge voltage data, compared to other references that use approximately 6000 seconds of data. Using the Oxford Battery Dataset, we show that FNN excels in early-cycle estimations, while GPR performs better over full cycles. After detecting and excluding poorly performing cells, the combined model achieves superior accuracy, highlighting its potential for second-life battery applications and cell balancing strategies.