• Optimized and deployed a TensorRT-accelerated inference engine for fine-tuned SLMs, achieving a significant reduction in latency and increasing throughput by 2-3x compared to standard CUDA-based implementations for NLP tasks such as hate speech and crime detection. Achieved 85%+ Precision and 90%+ Recall. Models handled millions of requests daily with low latency and minimal memory footprint across social media platforms.

• Engineered a high-throughput text processing pipeline for social media data at scale, deployed in a microservice-based architecture for efficient batch processing and real-time response.

• Integrated and automated human-in-the-loop data labeling systems, reducing manual effort and accelerating dataset curation for supervised learning.

• Built the ROS Navigation Stack for a Holonomic Drive Mobile Manipulator. Integrated and developed its Localization, Depth and Pose Sensing using T265 and D435 RealSense Cameras and SLAM Algorithm using GMapping.

• Built ROS Services and Actions for giving custom Goal Locations and used YOLO and MediaPipe Models for detecting Pose of Human Hand for perfect handoff of tools using UR5 Arm.

• Implemented a Communication System between the Mobile Base and the Hand Manipulator to enhance its Autonomous Capabilities.

• Demonstrated proficiency in working on Model Optimization for Computer Vision Tasks for Real- Time inference from sports videos.Implemented quantization that reduced the model size by 0.5x and reduced the latency by 3x

• Worked on large-scale image and text data using distributed data pipeline of PySpark and PostgreSQL Database. Wrote scripts for augmenting the database and Data Engineering.

• Worked on applying GCP based AI APIs like DocAI and Vision OCR

that was used to read up documents and extract the required data that

was further processed for our usage.

• Created a reboust accuracy script using pandas, numpy and other

related libraries that was designed according to our model.

• Worked on processing data in the GCP Cloud PostgreSQL Server.

• Data collection, cleaning, and preprocessing on the given sheet documents using Python.

• Delivering the model in a form that can be easily deployable and

maintained.

• Tested the end to end pipeline for the entire Machine Learning Workflow.