At Tecdatum Infoservices Pvt. Ltd., I engineered a Python pipeline and Docker-based deployment workflow for an automated vision inference system on edge hardware. It processed 10,000+ daily inference runs with automated telemetry tracking.
I also built a TensorRT optimization pipeline to compile, quantize, and benchmark deep-learning models. By analyzing latency-accuracy trade-offs, I reduced inference latency by 68% while maintaining 95% accuracy.
In my Multimodal Vision RAG Engine and Triton Real-Time Engine projects, I built retrieval and model-serving pipelines for multimodal AI and high-throughput inference. The projects reduced retrieval latency by 42% and inference latency by 3.2x, respectively.
