At CMS Computers Pvt. Ltd., I developed a YOLOv7 road-infrastructure defect detection model that identifies potholes, cracks, and other defects across 50 classes with 85% accuracy. I also used MLflow for model versioning and experiment tracking.
I designed and optimized a real-time video inference pipeline using NVIDIA DeepStream, TensorRT, and YOLOv7. It includes GPU-accelerated decoding, multi-object tracking, GPS-based event localization, and spatiotemporal deduplication.
I developed FastAPI services with SQLAlchemy and PostgreSQL for road-defect event management and dashboard visualization, and containerized applications with Docker. I also developed and deployed an enterprise chatbot using Llama 3.1 Instruct, FAISS, and vLLM, with hybrid retrieval, conversational memory, and RAGAS evaluation.
Previously, at Maximus Infoware Pvt. Ltd., I worked on ATM cash forecasting with LSTM and Transformer models. At Instacad Solutions Pvt. Ltd., I developed object-detection pipelines for engineering drawings and a RAG model for engineering documents, deploying models with TensorRT.
