Dishant Kumbhar

GCP-Kubeflow Migration: Orchestrated the seamless migration of a client from Metaflow with an airflow backened to kubeflow on vertex AI Platform, transforming a POC into a revenue generating project that yielded $200,000. Risk Prediction: Streamlined workflows, optimized and compared multiple pipeline runs & hyperparameters for significantly faster model development & improved risk prediction in sales & rentals with best Vertex AI practices. Empowered Talent: Successfully mentored & led a team of 3 through diverse POCs, fostering growth and maximizing team impact. Leveraged Langchain and RAGs to optimize LLM models for document summarization and question answering. Automated Loan Sanctioning: Designed Gemini-powered solution utilizing Langchain, RAGs, and Streamlit to automate data parsing, optimize credit assessments, and expedite loan approvals.

Developed an automated ML reporting tool (for PCL-RSHN): Optimized user experience by allowing them to upload multiple set of ground truth and prediction dataset. Implemented the capability allowing user-defined custom metrics templates to generate a customized report. Enabled user query to view the sub-group data and report can be downloaded. Research and Development: Conducted research on NLP models in healthcare domain like Clinical-Bert, Clinical-XLNet, Bio-Med Roberta, etc. Researched and implemented different Model interpretations techniques like SHAP, LIME, Saliency map, Grad-Cam, etc. Developed summarizer model using Document AI pipeline to summarize research paper with user defined word limit. Conducted Data analytics and prepared storyline, potential multiple business use case from various datasets having thousands of patients data. Chatbot / Voicebot Development: Developed a chatbot/voice-bot for a blood bank using Dialogflow, Node.js and webhook boilerplate. NLP- NBME Patient Score: Used Roberta, Deberta model to automate locating the features in patient score to evaluate candidate performance in the medical exam, improved the model performance using random-search from 78% to 83%.

Pune