Experience

1. September 2025 – Present, New York, NY

   Adobe

   AI Researcher

   • Building a personalized multimodal LLM pipeline for image editing using LLaVA, GPT-4o-Vision, and RAP (Retrieval-Augmented Personalization), leveraging Reddit PSR data, synthetic generation, and an agentic prompt-enhancement workflow.
   • Modeling user personas in latent space from edit deltas and linguistic patterns to enable persona-aware, instruction-faithful, and style-consistent edits.
2. May 2025 – Aug 2025, San Diego, CA

   Intuit

   AI Science Intern

   • Engineered a personalized tax topic recommendation system for TurboTax using Quantile Regression to predict time spent, reducing abandonment by 32%, lifting CTR by 60%, and increasing engagement by 45%.
   • Distilled GPT-4.1 as teacher into a LLaMA-3.1-8B student via reflection-based training for tax-field extraction, improving alignment on complex tax fields from 56% to 83%, increasing recall by 52%, and reducing inference cost by 89%.
3. Jan 2025 – Present, New York, NY

   Columbia University & University of North Carolina, Chapel Hill

   Graduate Research Assistant – Robert N. Butler Columbia Aging Center

   • Built a RAG-based biomedical Q&A system using GPT-4o, LangChain, and vector-based retrieval over research publications, achieving 81.2% top-3 document recall.
   • Developed an ASR pipeline using the Whisper model to transcribe Add Health study audio, achieving 87.4% precision and 82.1% recall on cognitive assessment tasks.
4. Jan 2023 – Jul 2024, Bengaluru, IN

   Hewlett-Packard Enterprise

   Software Engineer I

   • Developed a predictive maintenance microservice for HPE Proliant servers, decreasing downtime by 20%.
   • Implemented Autoencoders and LSTMs for server temperature data modeling, improving anomaly detection accuracy.
   • Built a backend framework automating log-to-JIRA defect mapping using spaCy, FastAPI, and ReactJS, cutting turnaround time from 3 days to 1 day.
   • Streamlined log aggregation via Kafka and Grafana, reducing redundancy fourfold and issue resolution time by 35%.
5. Jun 2022 – Aug 2022, Mumbai, IN

   Reliance Jio

   Data Science Intern

   • Performed exception reporting of network performance KPIs for 12M+ international roaming subscribers, flagging deviations above 10%.
   • Optimized event creation cycles from 45 to 15 minutes via Spark jobs, cutting reporting time by 67%.