Apr 2024 – Present · 2 yrs 2 mos · Bengaluru, Karnataka, India

• 🩺 Medical Document Standardization (MD-DAS)
• > Pipeline Development: Designed and implemented a type-agnostic processing pipeline to expand digitization and standardization efforts from lab reports to all medical documents.
• > LLM Training & Upstreaming: Managed the creation of training datasets, distillation, and RL training to upstream medical document data to Gemma model variants. Implemented comprehensive evaluation pipelines.
• > Terminology Prediction: Significantly improved clinical terminology prediction (standardization to FHIR) from 50% to >90%. Currently researching scalable methods (RAG, small LLMs, heuristics) for efficient LOINC/SNOMED code prediction.
• 🏥 Google Health Toolkit (Python, Kotlin)
• > LLM Agent & Production: Enhanced the LLM agent using Gemini and Gemma (text/multimodal) for efficient medical lab report digitization to structured JSON. Developed production expertise with Gemini (disaggregation, TPU optimization, deployment).
• > Prompt Engineering & Fine-Tuning: Demonstrated expertise in prompt engineering and fine-tuning Gemini/Gemma models for data digitization.
• > Mobile Health SDK & AI: Architected and developed the Kotlin Android SDK with an on-device FHIR database (sql-on-fhir-v2). Explored running on-device Gemini/Gemma models using ML Kit on Android and developed the essential backend API.
• 🌐 Open Health Stack (ohs.dev) Ecosystem Enablement
• > SDK Development & Impact: Developed a specialized Android sensing SDK for offline sensor data capture in low-resource environments. Provided technical guidance to partners like KhushiBaby, enabling them to leverage the Android FHIR SDK and touch 50 million lives.
• > Codebase Optimization: Directly contributed to the Android FHIR SDK codebase on GitHub, resolving critical issues and dramatically improving on-device search speed from 30 seconds to 30 milliseconds.
• > Community: Fostered community growth through proactive code reviews and active participation in developer calls.

Jun 2022 – Mar 2024 · 1 yr 9 mos · Bengaluru, Karnataka, India

• ABDM Wrapper Interoperability: Contributed to the development of the ABDM Wrapper using Java, facilitating seamless integration of legacy healthcare systems with the Ayushman Bharat Digital Mission framework.
• Roshnee Backend Development: Played a key role in building the backend for the Roshnee mobile-first platform using Java and Google Cloud Platform (GCP services), contributing to its capabilities in care collaboration and population health management.