The SPOTLIGHT project is a commensal survey for Fast Radio Bursts (FRBs) and pulsars at the uGMRT using real-time HPC and AI for simultaneous time-domain detection and arc-second imaging localisation across 300 to 1460 MHz. FRBs are searched within 2,000 post-correlation beams steered across the Field of View (FoV), with SPOTLIGHT leveraging its PetaFlops of computing power to process PetaBytes of data per day in real time. The AI-powered trigger for genuine astronomical signals enables capturing voltage samples spanning the burst duration to study the spectro-temporal-polarimetric properties at microsecond time resolution as well as high-time resolution visibility for event localisation. With an unprecedented sensitivity, SPOTLIGHT is expected to find hundreds of localised FRBs. This paper presents the FRB candidate optimiser for SPOTLIGHT's FRB detection pipeline, which incorporates candidate clustering, classification, and triggering for visibility and raw voltage recording.