Lakshya Singhal

Research Scientist @Frontera

Denver, CO, US

Built and deployed Person Re-Identification (ReID) systems with integrated object tracking for accurate multi-frame identity matching in videos- Applied Vision-Language Models (VLMs) like Qwen-VL and LLaVA for video summarization, action - recognition, and semantic video understanding- Fine-tuned VLMs using Reinforcement Learning (RL) techniques such as GRPO and PPO to enhance multimodal alignment and downstream task performance- Utilized LoRA and QLoRA for efficient parameter-efficient fine-tuning of large-scale multimodal transformer models