Rajaharsha  ·  Senior LLM / ML Engineer  ·  6+ yrs

Overview: Built an LLM pipeline to streamline the analysis and comprehension of research papers, facilitating easier access to medical research insights. Responsibilities: Handled and processed large text datasets efficiently, implementing pre-processing techniques like tokenization, stemming, and lemmatization. Established task-specific workflows using LLMs, utilizing orchestration frameworks like LangChain and LlamaIndex.

Overview: This solution was designed to save man-hours and reduce lead time in generating results for understanding crystallization processes. Responsibilities: Developed instance segmentation models for monochrome images, applying SSL and N-shot learning for domain adaptation. Utilized AWS Sagemaker and S3FS for robust model training and testing.

• This solution provides auto assessment of meibomian glands to quantify dry-eye syndrome, automating a crucial diagnostic step.

• Pre-processed datasets to remove noisy annotations and bias with pandas, numpy, and OpenCV.
• Customized segmentation models with pytorch, TIMM, MMSegmentation, and Tensorflow.
• Utilized GCP VertexAI for data management and training.
• Co-authored and published findings in a Data Science Consortium.

• Deployed all-purpose segmentation models suitable for a variety of datasets, improving model versatility and reusability.

• Investigated Meta's SAM on various datasets for few-shot performance.
• Ported segmentation models to Docker images using FastAPI, Shell, Pytorch, and Tensorflow.
• Deployed models for diverse datasets in JFrog and Kubernetes.
• Integrated image preprocessing methods as APIs with numpy, pandas, OpenCV, and Tensorflow.

• Built a visualization dashboard pipeline for gait analysis, providing real-time insights from sensor data.

• Developed a data processing and storage pipeline using AWS Kinesis and S3 for sensor data.
• Enhanced IMU fusion algorithm using Python for wearable device sensory data.
• Built an MLP framework for data training with CLI, including preprocessing and visualization.
• Established a real-time dashboard in Tableau with anomaly detection algorithms.