Vishesh Kumar Gautam

Directed end-to-end ML pipeline for a passive and lightweight face liveness detection project, encompassing data acquisition, model development, and rigorous validation phases.

Architected a comprehensive system, integrating advanced computer vision techniques for optimized dataset design, efficient preprocessing, and precise face extraction.

Developed and fine-tuned a CNN-based model, significantly improving the detection of spoofing attacks with high accuracy across diverse environmental conditions.

Containerized the entire training and inference pipeline using Docker, deploying the solution on RunPod to achieve scalable, GPU-accelerated execution.

Conducted in-depth research on state-of-the-art face liveness detection and anti-spoofing techniques, extracting key methodologies to enhance model robustness and overall detection accuracy.

Engineered and deployed a memory-aware AI avatar workflow, utilizing both long-term and short-term memory to significantly enhance contextual reasoning capabilities.

Executed a comprehensive comparative analysis of generative AI platforms (e.g., HeyGen, ElevenLabs), producing technical documentation on rate limits, token costs, and integration feasibility for key stakeholders.

Seamlessly integrated critical external APIs into the AI avatar pipeline and implemented robust error-handling strategies, ensuring high availability and system robustness.