Contributed to the development and extension of GPU-accelerated Python inference services using FastAPI,supporting transformer-based models for internal reference and customer validation workloads.

• Worked on Retrieval-Augmented Generation (RAG) prototypes using LangChain for orchestration and

LlamaIndex for document ingestion, focused on internal benchmarking and customer proof-of-concept

demonstrations.

• Evaluated vector retrieval approaches using FAISS (GPU) for internal experimentation and supported

Pinecone-based setups exclusively for external customer PoCs, maintaining clear separation between tooling.

• Applied parameter-efficient fine-tuning techniques (LoRA / QLoRA) during controlled experimentation using Hugging Face Transformers, leveraging pre-trained checkpoints and existing GPU infrastructure.

• Supported data preprocessing and evaluation workflows using Pandas, NumPy, and SQL to enable structured experimentation, offline evaluation, and performance comparison across model variants.

• Assisted in optimizing inference workflows through ONNX export and collaboration on TensorRT

benchmarking, observing improvements in latency and throughput.

• Containerized ML services using Docker and collaborated with platform teams to deploy GPU workloads on Kubernetes, emphasizing scalability and efficient resource utilization.

• Used MLflow for experiment tracking and model version comparison within prototyping and benchmarkingpipelines to support reproducibility.

• Supported customer-facing validation workloads on AWS (EC2 GPU instances, EKS, S3, ECR), assisting with deployment verification, benchmarking, and technical demonstrations.

• Participated in internal model evaluation discussions, contributing analysis on LLM behavior, hallucination patterns, and grounding quality in RAG systems under senior guidance

• Developed and deployed end-to-end machine learning pipelines using Python, PyTorch, and TensorFlow for enterprise-facing applications.

• Built supervised learning models (classification and regression) on large, structured datasets, improving

prediction accuracy by 10–15% through feature engineering and iterative experimentation.

• Performed data preprocessing, feature engineering, and EDA using Pandas, NumPy, and Scikit-learn,

enabling stable and reusable training pipelines.

• Integrated ML models into production systems in collaboration with software engineering teams, following internal deployment and validation standards.

• Deployed and monitored models using Azure Machine Learning, leveraging cloud-based compute for

training and batch inference workloads.

• Conducted hyperparameter tuning and A/B testing to improve model performance, stability, and inference efficiency.

• Used Jupyter Notebooks and visualization tools to communicate experimental results and insights to

technical and business stakeholders.

• Worked within Agile/Scrum development processes, participating in sprint planning, backlog grooming, daily stand-ups, and retrospectives to deliver incremental ML features.