Shyam Pawar Email and Phone Number

Drawing from my background at IIT Bombay, I developed and put into action a solution that leveraged Apache Kafka to enhance our company's video analytics capabilities across edge platforms. This solution facilitated the effortless transfer of video data to NVIDIA DeepStream cloud environments, making real-time video analysis faster and more efficient without any delays. Additionally, I developed analytics focused on specific areas, such as identifying outer-boundary intrusions and tracking how crowded certain areas are, to enhance the observation of people's movements, the tracking of items, and how long people are interacting with certain equipment, among other capabilities. These capabilities automated the need for constant manual monitoring, enhanced security with instant notifications, and optimized traffic management by allocating space and resources more effectively. To present these analytics, I created an interactive Svelte-based web app using Apache ECharts and OpenLayers. The platform provided a comprehensive dashboard for visualizing heatmaps, analyzing dwell times, tracking movement patterns, and assessing occupancy trends via an interactive floor map. Managers could easily transition from live tracking to reviewing past data, achieving sophisticated control and adaptability for making well-informed choices. Behind the application's user interface, I built a FastAPI server to handle the collection and processing of live tracking and analytics data from high-volume data streams. I ensured the seamless operation of the front-end interface, with the data being stored in TimescaleDB. Moreover, I implemented a WebSocket handler for managing dynamic subscriptions, adeptly handling incoming messages, updating analytics, and tracking data based on user-specified parameters. This strategy guaranteed that users received the most relevant and up-to-date information, thereby boosting the system's efficiency and effectiveness.

My career encompassed various cutting-edge technical areas, such as creating conversational agents through the Retrieval Augmented Generation (RAG) framework. This project aimed to make knowledge retrieval more efficient while maintaining the security and confidentiality of the information. It required the uploading of PDFs, their organization into smaller, more manageable sections, and their storage in ChromaDB. This configuration facilitated effective knowledge retrieval, enabling the LLAMA2, a large language model, to produce precise and contextually fitting answers. Moreover, PostgreSQL was improved with SQLAlchemy, scalable FastAPI interfaces were developed, PyTorch was set up for GPU acceleration in a Docker container, and Nginx was upgraded for enhanced scalability and performance.However, the standout achievements of my career were in the field of computer vision, where I led a project that utilized NVIDIA's DeepStream SDK for intelligent video analytics (IVA). This project involved enhancing the flexibility of the pipeline and integrating Quantized YOLO-NAS, -v8, and -v7 models for sophisticated object detection and pose estimation, to improve the system's capability to analyze and comprehend visual information. The DeepStream-based pipeline successfully tracked and counted people within a given area, visualizing the density of the population on floor maps by applying conventional image processing methods alongside DeepStream. This method offered valuable insights into the distribution of space. The system was also expanded to manage real-time video streams from up to 15 surveillance devices on a single GPU server, optimizing for high-speed processing and deploying models through Nvidia’s Triton Inference Server.

- Implemented Single Sign-On (SSO) from scratch with JWT Authentication & Authorization.- Made Dynamic Table with features like live editing, filtering, reordering, xlsv/csv exporting.- Implemented reusable code with code splitting.- Followed design patterns with good code practices.