Experience and mentors

The labs I’ve worked in, and the people who taught me how to do research.

Research experience

  1. July 2025 to present

    Student Researcher

    MIT CSAIL, supervised by Jehanzeb Mirza and advised by James Glass

    Test-time improvement, long-context reasoning and architectural advances for vision-language models.

    • Built Test-Time Reinforcement Learning for VLMs, which lets models self-improve without ground-truth labels through adaptive inference.
    • Designed frequency-based and diversity-based rewards that beat standard majority voting for test-time adaptation.
    • Evaluated on 15+ multimodal benchmarks, with strong gains even from a single unlabeled test sample and open-source VLMs surpassing GPT-4o by 2%.
    • Studied when and how models self-evolve in multimodal settings.
    • First-author paper accepted to CVPR 2026.
  2. May 2024 to July 2024

    Deep Learning Researcher

    HyperVerge, Bangalore

    Vision-language models and detectors for unstructured identity documents.

    • Used vision-language models to generate labelled data for unstructured identity documents.
    • Trained object detectors such as YOLOv8 for ID cards including voter ID and Aadhaar cards.
    • Built task-specific datasets for different ID types and applications.
  3. September 2023 to May 2024

    Student Researcher

    Multimodal Interaction Lab, IIT Jodhpur

    Dense 3D shape matching for clothed human bodies.

    • Converted triangular meshes into Möbius grid spheres, turning non-rigid shapes into rigid ones for simpler mapping.
    • Processed the transformed data with graph neural networks and spherical harmonics.
    • Used functional-map solvers for accurate shape matching; paper submitted to IEEE Signal Processing Letters.
  4. August 2023 to June 2024

    Student Researcher

    Multimodal Interaction Lab and DRDO, IIT Jodhpur, supervised by Rajendra Nagar

    LiDAR SLAM with neural distance fields, published as CCNDF.

    • Modelled point clouds as implicit surfaces with neural distance fields to map 3D surroundings.
    • Used the second-order derivatives of the field to compute the radius of curvature and constrain learning on complex geometry.
    • Improved mapping and localization, and explored RGB-D and LiDAR SLAM methods.
    • First-author paper accepted as an oral at ACCV 2024.
  5. June 2023 to May 2025

    Student Researcher

    Multimodal Interaction Lab and DRDO, IIT Jodhpur

    Spatio-temporal representations of dynamic scenes for autonomous vehicles.

    • Extended signed distance fields to real-time use with LiDAR point clouds.
    • Developed a method to segment static and dynamic objects in a moving scene.
    • Designed a geometric loss from distance-field gradients that regularizes the temporal domain, similar to eikonal regularization.
    • Improved on existing methods in surface reconstruction and dynamic-object segmentation.
  6. April 2023 to June 2024

    Research Intern

    Texas State University, supervised by Heena Rathore

    Anomaly detection for connected vehicles, and reinforcement learning under moral uncertainty.

    • Built an anomaly detection and recovery framework using Bayesian online changepoint detection and Bayesian forecasting.
    • Simulated sensor faults including white and Lévy noise, bias and spikes to test robustness.
    • Created a dataset from U.S. DOT Tampa connected-vehicle data; first-author paper in Elsevier Vehicular Communications.
    • Designed a voting-based system to combine divergent ethical theories in AI decision-making, with a Bayesian framework that updates moral beliefs from environment rewards.
  7. May 2022 to July 2023

    Student Researcher

    Multimodal Interaction Lab, IIT Jodhpur

    Text-to-image generation and cultural bias.

    • Worked on neural text-to-image generation with VAEs and transformers.
    • Created the Indian context Image caption Dataset and Indian Heritage Dataset, with 250k+ image-caption pairs.
    • Sole-author student abstract accepted at AAAI 2024.

Mentors

Rajendra Nagar

Supervisor, IIT Jodhpur

Heena Rathore

Advisor, Texas State University

M. Jehanzeb Mirza

Advisor, MIT CSAIL

Wei Lin

Advisor, JKU Linz

I’ve also been fortunate to work with some amazing colleagues — Shyam Marjit, Deepanshu Solanki, Karan Bhakuni, and Paul Gavrikov.

Summer schools

Amazon ML Summer School, 2024

Selected among the top 1,000 students across India.

Climate Change AI Summer School, 2023

Computer vision for remote tracking and monitoring of ecosystems.