Cv

MS student at UC San Diego working on computer vision, graphics, and vision-based agentic AI.

Basics

Name Gaurav Joshi
Label Computer Vision, Graphics & Agentic AI
Email gajoshi@ucsd.edu
Url https://gaurav17joshi.github.io/
Summary My research sits at the intersection of computer vision and agentic AI, focusing on how models perceive, reason about, and act within visual and digital environments. This includes work on inverse graphics, procedural content generation, and computer-use agents that translate visual and expert knowledge into executable actions.

Education

  • 2025.09 - Present

    San Diego, CA, USA

    MS in Computer Science
    University of California, San Diego
    Computer Vision & Graphics, ML & AI
    • Advisor: Manmohan Chandraker
    • GPA: 3.95/4
  • 2021.11 - 2025.05

    Gujarat, India

    B.Tech in Computer Science & Engineering
    Indian Institute of Technology Gandhinagar
    Computer Science & Engineering
    • Advisor: Shanmuganathan Raman
    • GPA: 3.96/4, Dean's List

Work

  • 2025.10 - 2026.03

    San Diego, CA, USA

    Research Assistant
    UC San Diego – Center for Visual Computing
    Advisors: Manmohan Chandraker (Prof., UCSD), Kunal Gupta (PhD, UCSD)
    • Conducted research on LLM-based procedural material generation for Blender using expert workflow reasoning; built automated pipelines to extract structured procedural knowledge from Blender tutorial videos via video understanding and LLM-based analysis.
    • Developed retrieval-time reasoning methods that synthesize expert-inspired material creation processes before compiling them into executable Blender node graphs.
    • Evaluated the system through expert and large-scale user studies, demonstrating improved workflow quality over graph-only procedural generation methods.
    • Publication: "Reflecting Process Expertise in Procedural Material Generation", accepted at ECCV 2026.
  • 2024.09 - 2025.06

    Gujarat, India

    ML Intern
    IIT Gandhinagar – CVIG Lab
    Advisors: Shanmuganathan Raman (Prof., IITGn), Ashish Tiwari (PhD, IITGn)
    • Worked on inverse modeling for outdoor scenes from single images.
    • Used Gaussian Splatting and NeRF models to represent outdoor scenes and perform reconstruction/re-rendering.
  • 2024.05 - 2024.07

    West Lafayette, IN, USA

    Research Intern
    Purdue University – IDEAS Lab
    Advisors: Aniket Bera (Associate Prof., Purdue), Hrishikesh Vishwanath (PhD, Purdue)
    • Worked on operator learning (neural operators) to build surrogates for fluid simulation (Navier-Stokes equations).
    • Worked on human-pair motion generation using generative neural operators.
  • 2023.05 - 2023.08

    Remote

    GSoC Contributor
    Google Summer of Code — OpenAstronomy / Stingray
    Advisors: Daniela Huppenkothen (Univ. of Amsterdam), Matteo Bachetti (INAF, Italy)
    • Developed a Gaussian Processes inference tool to calculate the evidence of a QPO (Quasi-Periodic Oscillation) signal in astronomical time series from events like GRBs and magnetar/solar flares.
    • Made demonstration and tutorial notebooks on Bayesian inference and the project tool.
    • Used Jax, TinyGP, JaxNS, and tensorflow_probability.
    • Open-source project funded under the Google Summer of Code program.

Publications

Projects

  • 2023.10 - 2023.12
    PDEs Solving using Physics-Informed Neural Networks
    Self project
    • Implemented PINNs to solve Laplace equations for various Dirichlet conditions.
    • Comparative evaluation against analytic and mesh methods.
    • Tools: TensorFlow, LaTeX.
  • 2023.07 - 2023.10
    Computational Statistics eBook
    Jupyter Book
    • Wrote an online book on computational statistics covering the math behind Bayesian data analysis and parameter estimation.
    • Covered random variable generation, Monte Carlo integration, MCMC (MH, HMC, Gibbs), Gaussian processes, information theory, sequential Monte Carlo, and graphical models.
  • 2023.07 - 2023.08
    Effective Sensor Placement using Neural Processes
    Mentor: Prof. Nipun Batra
    • Built an ML pipeline using conditional neural processes for effective pollution sensor placement.
    • Used CAMx-simulated pollution data to predict future pollution levels and place sensors via a mutual information criterion.
    • Compared fit quality across neural processes, Gaussian processes, and other models.
    • Tools: PyTorch, Jax, DeepSensor.
  • 2023.03 - 2023.05
    Scientific Computing eBook
    Julia eBook
    • Wrote an online book on scientific computing with Julia, focused on scientific machine learning.

Awards

Skills

Research Areas
Computer Vision
Computer Graphics
Agentic AI
Probabilistic / Bayesian ML
Deep Learning
High-Performance Computing
Numerical Analysis
Programming Languages
Python
JAX
C
C++
Julia
Java
Tools & Other
Git / GitHub
Jupyter Notebooks
Google Colab
Unix
LaTeX
Jekyll
HTML