Subhadeep Mandal Email & Phone Number

Bengaluru, Ka, In

Bengaluru, Karnataka, India

- Deep Learning and Generative AI R&D in Healthcare and Finance: Generated data synthetically using proprietary and open-source LLMs for fine-tuning LLMs and VLMs for finance, insurance and medical use-cases. Fine-tuned Phi 3 and Phi 3.5 using the synthetically generated datasets. Used Unsloth for faster and memory efficient fine-tuning. Evaluated the models on test-set and created comprehensive report for reference.- Multimodal Retrieval-Augmented Generation for QA Systems: Build multi-modal RAG application for question answering on documents (scanned PDFs).- Enterprise GenAI and MLOps Solutions in Healthcare and Finance: Deployed fine-tuned models on cloud services. Built consumable REST APIs serving endpoints to run inference on the deployed models.- State-of-the-Art Audio Language Frameworks Implementation: Built end-to-end plug-and-play framework and application for audio based use-cases involving summarization, sentiment analysis, insights and feedback generation. - Developing Low Code Frameworks for Large Vision Models Fine-Tuning: Developed a library for low code fine-tuning traditional computer vision models using custom data, with minimal code. Supports image classification, segmentation, object detection, image generation and VLM fine-tuning.- Papers and patents authoring: Authored in two papers over audio use-cases for IIMB ICBAI-2024 conference. Currently focused on patenting our latest RnD solutions on Generative and multi-modal AI.

Bengaluru, Karnataka, India

– Developed state-of-the-art vision transformers for medical and industrial applications. Created explainable AI solutions to visualise and interpret model performance.– Designed and built a Flask web application with an HTML user interface to efficiently utilise the model and provide real-time results. Conducted experiments with MONAI Deploy SDK for model deployment in the medical domain.– Generated a comprehensive report comparing models based on key metrics such as performance, accuracy, training time, and inference time.– Explored vision-language models tailored for medical applications. - Developed smart agent based solution with Langchain to automate Doc QnA, Source Retrieval and Summarization with RAG, as a part of Genpact's Inhouse Hackathon.- Explored few Multimodal VLMs for Visual Question Answering, focusing medical domain.- Developed model for detecting AI generated text. Developed Gradio and FastAPI based User Interface utilizing the models.- Experimented with Nvidia's RIVA framework and OOTB models for S2T, T2S and NMT tasks. Created a comprehensive report comparing performance of Nvidia's and open source audio models.

Bachelor Of Technology - Btech, Computer Science, 9.04

Higher Secondary, Pcm, A

Secondary School, A+