Manish Sharma

I’m an AI Engineer with 5.5+ years of experience in Generative AI, diffusion-based text-to-video pipelines, and multimodal robotics. I build systems that connect vision, language, and video into practical, scalable capabilities—especially where simulation and real-world performance must align.

In my current role as Tech AI Lead at Alphadroid, I designed video generation pipelines using diffusion transformers and spatio-temporal attention, producing 500K+ diverse training clips. This work improved robotic manipulation robustness by 67%, and it also enabled Sim2Real training frameworks that reduced physical data collection needs by 70%.

I also integrated world model architectures (Dreamer, COSMOS-style) with NeRF and 3D Gaussian Splatting into navigation and perception modules, enabling predictive planning and context-aware behavior. On the efficiency side, I optimized video inference using CUDA kernel fusion, INT8 quantization, and pruning—cutting latency from 800ms to 120ms on edge devices.

Previously at Oodles Technologies, I led the shift from traditional ML to Generative AI by integrating GPT-4 and Claude into a Wellsite platform, reducing equipment failure prediction errors by 35% and preventing $2M in potential losses. I also built computer vision pipelines with OpenCV and YOLOv8 (96% detection accuracy), and delivered scalable multi-sensor data processing with Apache Kafka and Druid (20GB+ daily) while mentoring junior developers. I’m energized by building reliable ML systems with strong engineering standards, from pipelines and inference to production delivery.