Tushar Arvind Kamble

I’m a Senior Data Scientist who builds and ships AI systems—especially LLM fine-tuning, retrieval-augmented chatbots, and end-to-end ML automation. In my current role as Sr. Data Scientist - Engineer, I fine-tune GPT-4, LLaMA2-70B, and Mistral using LoRA/QLoRA/DoRA with LangChain and LlamaIndex, integrated with Azure AI Studio and Azure AI Foundry, and I’ve driven a 40% boost in ROUGE and BLEU with vLLM backends.

Previously at Ascendion, I spearheaded a machine learning team of 4 to engineer an automated recruitment system, strengthening candidate selection and improving recruitment efficiency. I also led ETL pipeline development with AWS Glue, Redshift, and Athena, improving feature selection from 300,000+ data points through EDA and Select K Best, with hyper-parameter tuning.

I focus on measurable impact in production—deploying forecasting models with Azure ML, Docker, MLflow, and observability on Azure Monitor, and accelerating rollouts using automated CI/CD. I’ve also built scalable application workflows with FastAPI and OpenAI on Azure Databricks, combining Kubernetes, Kubeflow, Spark, and PostgreSQL to reduce deployment time from 4 days to 2 minutes.

My work extends across applied research and product delivery: from chatbot systems using OpenAI and Llama2 with transformer tailoring (BERT, DistilBERT, CodeBERT) to multimodal RAG with CLIP embeddings and CI/CD for real-time responses. Earlier, at Tap Cube Studios, I developed telemetry and A/B testing infrastructure across 70+ games, and I continue to bring that same rigor to projects like HR document-processing chatbots and pneumonia detection using TensorFlow, YOLOv8, and OpenCV.