Experience

Research, engineering, and teaching roles across industry and academia.

  1. Applied Scientist Intern — Microsoft

    Redmond, WA May 2026 – Aug 2026

    • Worked with the Microsoft Copilot Studio (MCS) Team to improve and optimise the agent skills inside the MCS platform.
    • Developed a standalone evaluation scoring system to assess the performance of the Agent Skills.

  2. Summer Research Intern — Emergence AI

    New York City May 2025 – Oct 2025

    • Worked with a long-term memory system for conversational agents.
    • Achieved SOTA on LongMemEval with 86% accuracy, surpassing the previous best by 15%.
    • Developed specialised agents for enterprise tasks in planning and self-improvement.

  3. Graduate Research Assistant — UMBC

    Baltimore, MD Jan 2022 – Present

    • Developed ISM, a self-improving strategy memory for continual mathematical reasoning (ICML 2026 · AI4Math Workshop).
    • Enhancing mathematical reasoning in LLMs through human-driven learning techniques.
    • Designed Sim2Real framework with YOLO object detectors — 85% transfer success to real-world robots.
    • Enhanced nano drone navigation efficiency by 30% using Hierarchical RL.
    • RLHF via TAMER (Scalar binary feedback) for sequential decision-making agents.
    • Integrated graph attention networks in R-GCN — 20% improvement in sample efficiency.

  4. Graduate Assistant — UMBC

    Baltimore, MD Aug 2021 – Dec 2021

    • Assisted in teaching CMSC 691.4 (Intro to Data Science) for 50+ students, ensuring 95% assignment completion.
    • Collaborated with professor on student projects, increasing engagement rates by 20%.

  5. Software Engineer — Titan Company Limited

    Bengaluru, India Jul 2019 – Apr 2021

    • Built Spring Boot + REST API web app for 8,000+ employees — 40% reduction in data retrieval time.
    • Python dashboard for login analytics — 25% improvement in user engagement tracking.

  6. Software Engineer Intern — Titan Company Limited

    Bengaluru, India Mar 2019 – Jun 2019

    • Improved software reliability by 30%, reduced bug occurrences by 20%.
    • Tested BPM-related applications, increasing process efficiency by 25%.

  7. Research Intern — National University of Singapore

    Singapore Jun 2018 – Jul 2018

    • Achieved 98.5% accuracy on credit card fraud detection — outperformed ANN, RF, SVM.
    • Advanced preprocessing reduced data redundancy by 25% and training time by 30%.