Deepak Mangla

Architected Alethia AI’s real-time AI avatar platform, combining custom lipsync, emotive facial expressions, hand gesture synthesis, and low-latency streaming into a unified production system. Built autoscalable GPU infrastructure that benchmarked 200+ GPUs, reduced cloud costs ~20x, and enabled low-cost OME + SRT/WebRTC livestreaming supporting 40+ agents and 500+ concurrent users.

Architected the real-time AI avatar pipeline, including custom lipsync, emotive facial expressions, hand gesture synthesis, and seamless emotion transitions. Trained a custom lipsync model from scratch and reduced avatar latency from ~6s to <1s, while optimizing Flux image generation 3x across pipelines.

Founded a virtual try-on startup using computer vision for real-time garment visualization and secured partnerships with BookMyShow and FanCode/Shibumi.AI.

Led a computer vision team to automate NFT animation for major collections, processing ~200K animations and pioneering NFT dancing via keypoint-based human motion transfer (~12,000 unique dances). Deployed a generative art pipeline with multi-GPU support and fine-tuned diffusion models on LAION-5B while hiring and coordinating 10+ animators and engineers.

Optimized a First Order Motion (FOM) model for real-time face animation using super-resolution upsampling, PixelShuffle, and segmentation masks. Built a full lipsync pipeline and added a VQGAN-CLIP wrapper API with segmentation-based color shift and transparency support.

Built and deployed deep learning computer vision solutions for production use cases.

Built a generative AI product for real-time face-swap in videos, as an early mover in AI-driven creative content.