devchait Goto Github PK

• 👋 Hi, I’m @devchait
• 👀 I’m interested in software designing, AI & Neural Network, Computer Vision and building intelligent softwares
• 🌱 I’m currently learning NN-designs with pytorch and CUDA GPU Programming with C++17
• 💞️ I’m looking to open source some of my developments and continue my bloging on Python designing and learning Data-Science together effectively
• 📫 How to reach me [email protected]

Resume details:

[Your Name] [Your Address] [City, State, Zip Code] [Your Email Address] [Your Phone Number]

[Your Name] [Your Address] [Your City, State, Zip Code] [Your Email Address] [Your Phone Number]

Objective: Dedicated and highly skilled AI/ML Engineer with a strong background in designing and implementing cutting-edge solutions for deep learning inference. I have a proven track record of successfully architecting and developing frameworks for real-time video analytics, tracking, and server solutions. My expertise in leveraging multiple libraries and platforms, such as TensorRT, PyTorch, OpenVINO, and GStreamer, allows me to create scalable and efficient AI solutions. I am seeking challenging opportunities to contribute my skills and drive innovation in the field of AI/ML.

Projects:

Multi-faceted Framework for Benchmarking Deep Learning Inference Solution

Deployable Video Analytic solution serving 25-30 real-time rtsp-streams with 30 fps as analytical stream-out solution. 95% reduction in model deployment time and 85% reduction in development time for adding new business rules through custom plugins. ONVIF protocol support for analytical meta-data sharing. Multi-platform inferencing solution supporting OpenCV, OpenVino, PyTorch, and TensorRT for real-time benchmarking. Scalable on both GPU and CPU, supporting Windows and Ubuntu. Responsibilities: Architected and developed the entire framework from scratch, deployed the solution on client machines, benchmarked the performance of business rules across various frameworks, and maintained the GitHub repository for the solution. Design Patterns: Incorporated multiple design patterns like Facade, Bridge, Adapter, Proxy, Singleton, Observer, and Visitor. Libraries: RabbitMQ, PyTorch, TensorRT, Python bindings, OpenCV, OpenVino, Shapely, pytest. Time Taken: February 2020 - July 2020. Vicon AI: Mass Video Analytic Server Solution

Architected a server solution addressing 150-300 rtsp streams with 20-30 fps on Intel Xeon with 4 Tesla T4 cards. Enabled real-time communication of analytical stream and meta-data to multiple VMS nodes. Developed analytical library abstracting the complexity of analytics with a simple API for pipeline analytic stages. Implemented both CPU and GPU-based inferencing strategies to achieve low latency and high throughput. Led a team of 5-6 developers, maintained the GitHub repository, and implemented CI solutions. Platform: C++ 11 and 14, TensorRT, CUDA, Windows, Ubuntu, UnitTest, CMake 3.2, OpenVino. Design Patterns: Followed SOLID design patterns with a pipeline paradigm. Time Taken: August 2020 - December 2022. Real-time Tracking Framework

Templated framework in Vicon AI Server solution for real-time light-weight multi-object tracking for mass streams. Achieved low latency and encapsulated object information as Frame State meta-information. Designed and implemented tracking framework with plugin mechanism for easy deployment of new Association algorithms. Implemented both CPU and CUDA-based Association Algorithms with significant performance gains. Optimizations: Implemented faster custom CUDA kernels for YOLO Pre-processing and modified YOLO CUDA kernel for improved tracking performance. Scalability: Utilized GStreamer, DeepStream, and Triton for large-scale inference servers. Time Taken: January 2023 - Ongoing. Hands-On Experience:

PyTorch Modeling: Conducted research and implemented various models, including customized DataLoader for custom datasets. Model Deployment: Trained and deployed PyTorch classification models, experienced in using various loss functions. Model Optimization: Converted ONNX models to TensorRT engines and serialized PyTorch weight files to TensorRT engines. Custom CUDA Kernels: Developed custom CUDA kernels for YOLO Pre-processing and NMS plugin layer in TensorRT. API Development: Exposed API functions for Python bindings and separate bindings for CUDA kernels.

DataLablePro:

Role: Product Manager and Solution Architect. Responsibilities: Ideated new features, designed interactions, led the implementation, and maintained the backend ML library. Key Contributions: Structured the solution as a multi-faceted architecture to accommodate independent implementation of loose scripts, leading to a versatile product. ML Operations: Involved in data set labeling to model benchmarking and contributed to the ecosystem's architecture. Key Features: Developed queue-based annotation task assignment, image and video annotation review and feedback, annotation format converter, model accuracy validation, version association for AI models and datasets, project progress tracking, and contribution insights. Platform: GitHub, DVC, Python bindings, PostgresSQL, Micro-Services, ReactJS, Node, MOT benchmarking, COCO, plugins. Time Taken: January 2021 - Ongoing. Education: [Bachelor's/Master's Degree in Computer Science or related field] [University Name], [Year of Graduation]

Skills:

Deep Learning: PyTorch, TensorRT, OpenVINO Programming Languages: C++, Python Libraries & Frameworks: OpenCV, GStreamer, CUDA Version Control: Git Web Technologies: ReactJS, Node.js Database: PostgresSQL Design Patterns: SOLID, Facade, Bridge, Adapter, Proxy, Singleton, Observer, Visitor Platforms: Windows, Ubuntu, Triton, DeepStream Certifications: [Optional: List any relevant certifications]

Languages:

English: Fluent (Native/Bilingual) [Other Languages]: [Level of Proficiency]