Experience

  1. Graduate Research Assistant
  2. R&D Data Science Intern, Generative AI
  3. Senior Analyst - Machine Learning Engineer
  4. Data Scientist
  5. Research Intern

Graduate Research Assistant

University of Massachusetts – Dartmouth, USA

Jan 2025 – Sep 2025

  • Developed a medical Visual Question Answering (VQA) pipeline by performing supervised fine-tuning on the AlphaCLIP + LLaVA-Med multimodal model (with 7.57 billion parameters), using the PEFT technique.

  • Trained a CLIP (Contrastive Language Image Pre-training) multi-modal model from scratch: developed vision and text encoders, integrated Byte Pair Encoding (BPE), applied contrastive loss, and deployed the model using Gradio.

  • Fine-tuned the Mistral-7B model on 75,000 financial news samples using the Parameter-Efficient Fine-Tuning (PEFT) technique, and utilized Weights & Biases for experiment tracking.

R&D Data Science Intern, Generative AI

Johnson & Johnson Innovative Medicine, USA

Jun 2024 – Aug 2024

  • Developed a conversational agent using LangChain, which incorporates retrieval-augmented generation (RAG), advanced prompt design, and memory-aware chain components.

  • Integrated Weaviate for vector search, with sematic routing and prompt-tuning.

  • Enhanced answer accuracy by 65% and cut response latency from days to <1 sec.

  • Delivered oral presentation at the 2024 J&J Global Intern Research Symposium—selected as 1 of 14 presenters among 85 interns.

Senior Analyst - Machine Learning Engineer

Tiger Analytics, India

Feb 2022 – Jul 2023

  • Designed a computer vision pipeline for oral video analysis using Azure ML + Databricks, slashing processing time by 70%.

  • Built a Label Studio-annotation pipeline, reducing manual effort by 65% and deployment via Azure ML.

Data Scientist

BlueReef Technology, India

Aug 2020 – Jan 2022

  • Developed an end-to-end Natural Language Processing (NLP) solution by integrating Amazon Web Services (AWS) such as S3, Lambda, Transcribe, and SageMaker to process over 100 hours of audio content, thereby automating text analysis and reducing processing time by 75%.

  • Implemented a sensor-based activity prediction model to process over 10 million data points collected from accelerometer and gyroscope sensors.

  • Designed advanced feature engineering techniques, including statistical measures and temporal pattern extraction.

Research Intern

Indian Space Research Organization (ISRO), India

May 2019 – Jul 2020

  • Developed PolInSAR techniques: Random Volume over Ground (RVoG) forward model + Three‑Stage Inversion model to estimate forest stand height.

  • Achieved a correlation coefficient of 0.81 and a root mean square error (RMSE) of 5.05 m for the Saipung Reserve Forest in northeastern India, located in Meghalaya.

  • Published work in Springer and presented at the 3rd conference of the Arabian Journal of Geosciences 2020, Sousse, Tunisia.