Shyamsundar Prabhakar Indra Email & Phone Number

Hyattsville, Md, Us

Maryland, United States

Summary : Specialized in MLOps & developing software for automated AI deployment on Renesas edge devices• Built a Python edge AI deployment software platform with end-to-end ML pipelines catered to 1D & 2D data from Renesas edge devices’ sensors, enhancing deployment speed by 46% and reducing manual work by about 70%, saving over 100 man-hrs/month.• Reduced the size of PyTorch Fully Connected Neural Networks by 42% through the implementation of Low-Rank Adaptation (LoRA), making models more efficient for edge deployment.• Integrated ONNX Runtime in the platform, improving model inference speed by 30% & reducing training time by 12%.• Developed and deployed a Support Vector Machine (SVM) model for voice authentication on a Renesas edge device, using a 500 samples in-house dataset collected manually, achieving 96% accuracy on 100 unseen people from the wild.• Dockerized the software platform, enabling consistent and scalable deployment across various developers and edge devices.

United States

Summary : Conducted cutting-edge research on generative AI models for text-to-3D facial animation• Performed comprehensive literature review on 3DGANs, GAN Inversion, & CLIP models for a novel text-to-3D animation generation methodology in the domain of 3D facial animations.• Solved an intermediate problem of text based 3DGAN image editing by creating a model integrating EG3D with GAN Inversion for text-based manipulation of latent space using CLIP, trained using the EmotiW (facial emotions) dataset.• Evaluated the intermediate approach using PSNR and SSIM scores to ensure similarity to the originally generated image and CLIP scores to ensure editing success in accordance with the textual input.

United States

Graduate Teaching Assistant for ENME272: Computer Aided Design

Bengaluru, Karnataka, India

Summary : 3D obstacle detection and lower level controller for WIPRO sponsored L3 autonomous vehicle (AV)• Developed LiDAR-based 3D object detection model for WIPRO sponsored L3 AV, achieving mAP values of 73% (cars) & 72% (pedestrians) on KITTI3D dataset, by training on Waymo Open dataset & self-curated Carla dataset of 10000 synthetic samples. • Engineered Control Barrier Functions (CBFs) based lower level controller python software package, which along with the 3D object detection, gave around 92% successful collision avoidance over 50 simulated scenarios in Carla. • Verified the controller by ROS deployment on a Copernicus UGV, resulting in a published conference paper on the same.

Bengaluru, Karnataka, India

Summary : ROS Perception stack and prototype development for autonomous farm robot• Curated & annotated around 2000 plant images using Azure Image Labeling tool for YOLO model training using transfer learning.• Achieved 94.6% accuracy on a publicly available leaf dataset, by training a YOLO leaf detection model for leaf counting, hence enhancing a farm robot’s crop health monitoring capabilities.• Led a team of 5 junior interns and accelerated completion of farm robot prototype development and ROS perception stack deployment with the leaf detection model on the same, within 22 days.

Pilani, Rajasthan, India

• Successfully defended thesis on Deep Learning applications in Computer Vision for level-3 autonomous vehicle.• Created a pedestrian detection + tracking pipeline in a multi-camera setup for an L2 AV, with open source models.• Engineered a Bi-LSTM based model to take tracking data as input and predict the path of pedestrians 2 seconds into the future.

Chennai, Tamil Nadu, India

• Served as a Summer Intern, leading the full-stack development of a dynamic web application. Handled back-end development using Java and JSP, while crafting an intuitive front-end with HTML and CSS.• Designed and implemented a user-friendly web interface for seamless input, storage, management, and retrieval of all project data for the institute from a MySQL database.

Master Of Engineering - Meng, Robotics, 4.00/4.00

Activities and Societies: - • Student athlete, Collegiate badminton teamCoursework: • CMSC720 - Foundations of Deep Learning • ENPM808Z.

Bachelor Of Engineering, Mechanical Engineering, 8.20/10.00

Activities and Societies: - • Student athlete -Track and Field team • President, PTM - Tamil Regional Association • Student Mechanical.

High School, Cbse Aissce - 97%

Cbse Aisse - 10.00/10.00