Work
My work has revolved around building AI systems, from production systems at scale to AI research organizations and foundation models for biology and medicine.
Journey · Tempus AI · Google · Talks & blogs · Publications · Patents · Education
Career journey
AI in biopharma R&D
Continuing to work at the intersection of AI engineering, research and biological science.
Tempus AI · VP, AI & Research
Oncology foundation models, clinical and research AI, training and inference infrastructure and building an AI organization.
Google · ML leadership
Production ML spanning conversational intelligence and safety systems, alongside engineering teams supporting global products.
Engineering foundations
Embedded systems, distributed systems, networking and large-scale software engineering.
Tempus AI
VP, AI & Research · Dec 2023 – 2026
Tempus applies AI to large-scale clinical, molecular and imaging data for diagnostics, research and therapeutic development.
I joined before the IPO as an AI leadership hire. My work included AI strategy, scaling the research and engineering organization, building a shared ML operating model and collaborating with technology and life science partners. I regularly presented AI strategy to the board.
I hired the initial team and built the AI research and engineering organization from the ground up. I established a hub-and-spoke AI operating model and developed a shared career ladder, performance review and calibration process. The organization worked across four areas:
Foundation models
Multimodal and longitudinal oncology foundation models, including a joint initiative with AstraZeneca and Pathos. Strategy, architecture, evaluation and research governance.
AI training and inference platform
Infrastructure for large-scale model training on H100 and GB200 GPU fleets, along with a scalable inference layer for open-source models processing millions of clinical documents.
Clinical AI
Clinical abstraction, trial matching, care gap detection and applications built on unstructured health records.
Biomedical research AI
AI-native tools and agentic workflows for biomedical research teams and pharma partners.
ML leader · Nov 2012 – Dec 2023
Over eleven years at Google I grew from individual contributor into engineering leadership across Core and Ads ML. I worked on applied ML, ML platforms, conversational AI and abuse prevention, building production systems and multi-team organizations. The work received 20+ organizational awards, including Google Feats of Engineering and Core Excellence.
Safety
ML models for abuse detection and policy enforcement across Ads, YouTube, Drive, Gmail and other products. See the 2022 and 2023 Ads Safety reports.
Conversation intelligence
Core platform for an AI-first omnichannel conversations product supporting 10M+ interactions per month. Work included summarization, speech quality, topic classification, smart replies and knowledge linking. Product context ↗
Talks & blogs
Publications7 research abstracts
AACR 2026:
- Agentic AI workflow for automated, high-fidelity curation of cancer diagnosis and staging from unstructured patient records · A hybrid, multi-agent AI workflow using a three-stage design — NLP pre-screening and LLM extraction — to autonomously extract cancer diagnoses and tumor characterization. Benchmarked against manual abstraction on 1,497 patients from the Tempus real-world database.
- Reasoning-guided retrieval improves oncology trial eligibility matching · An agentic retrieval strategy enabling LLMs to autonomously search and synthesize complex eligibility evidence from unstructured clinical notes. Benchmarked against a standard RAG system across 618 patient-level eligibility assessments and 148 biomarker extraction tasks.
ASCO 2025:
- Research · ASCO 2025Evaluation of LLM-based clinical abstraction of EHR for non-small cell lung cancer patients ↗ · Benchmarked LLM-based extraction of structured clinical data from EHRs against manual abstraction for NSCLC patients, evaluating accuracy across key oncology data elements.
- Querying a large oncology database with Generative AI ↗ · Developed and evaluated a natural language interface enabling clinicians and researchers to query a large real-world oncology database, reducing dependence on structured query languages.
- Patient records to timelines: A LLM-based approach for summarizing oncology patient history ↗ · Built and evaluated an LLM pipeline converting unstructured oncology patient records into structured longitudinal timelines, enabling downstream analysis and model training.
- Impact of large language models on trial-to-patient matching efficiency ↗ · Measured the impact of LLM-powered automation on matching patients to clinical trials, quantifying reduction in manual screening burden across a real-world oncology database.
- A NLP approach for OCR-resilient extraction and structuring of karyotype data in oncology clinical notes ↗ · Developed an NLP pipeline for extracting and structuring karyotype data from clinical notes, robust to OCR errors, enabling large-scale cytogenetic analysis in real-world records.
Patents6 patent entries
- Patent pending: Systems and methods for foundation model for oncology
- US20240404702A1 ↗ · Systems and Methods for Deploying a Machine-Learning Model for Performing a Specific Clinical Task
- WO2024249684A2 ↗ · Systems and methods for generating and deploying task-specific agents
- US20240428003A1 ↗ · Automatic content item updation based on computer-mediated interaction
- US20130083902 ↗ · Unique global identifier header for minimizing prank 911 calls
- US20130086274 ↗ · Timing management system for smarter decisions in time constrained distribution systems
Education
- University of Washington, Seattle · M.S. Computer Science and Engineering
- National Institute of Technology, India · B.S. Electrical Engineering · Merit scholarship