Sachin Venkatesh
01 / About

Sachin Venkatesh

I'm Sachin Venkatesh, a Physics PhD at Northwestern University, working broadly on physics-informed ML+AI advised by Prof. Alexander Tchekhovskoy. I was previously doing my PhD at Georgia Tech advised by Prof. Matthew Liska, where I received my MS in Computational Science and Engineering and MS in Physics. Prior to that, I received my B.Tech in Engineering Physics from Delhi Technological University.

My current research primarily focuses on physics-informed neural-operator emulators for 3D GRMHD simulations — forecasting full simulation states on the native grid as boundary conditions for a hybrid simulation+emulation framework. The goal is emphasizing physical validity like conservation laws and statistical rigor over pure pixel error. Previously, I worked on unsupervised GRB classification at NAOJ Japan, IllustrisTNG at CCA Flatiron, turbulence super-resolution at DTU, and as a Data Engineer (Snowflake/SQL/PySpark) at Genpact.

Core stack spans PyTorch/TensorFlow, JAX, CUDA, MPI, and multi-node accelerated training.

I'm always happy to talk and learn more about physics-informed ML, GRMHD astrophysics, Optimization, or time-series ML — please reach out below.

Industry Resume Academic CV
02 / Research Spotlight

Research spotlight

H-AMR GRMHD simulations vs. TRON emulator

Hybrid emulator for 3D GRMHD simulations — in prep.

Multiple snapshots (cadence Δt = 25 Rg/c), 8 primitives from a 3D GRMHD simulation + full Kerr metric. Factorized angular+radial convolutions, Fourier neural operator bottleneck, 9-objective physics loss trained with backpropagation through time. Stable 10/25/50/100-step rollouts preserving mass profiles, physical diagnostics, and jet morphology for hybrid simulation+emulation boundary conditions.

Ground truth vs 10/25/50/100-step density forecasts in xz and xy planes

Stability across horizons. Same target frame forecast from 10 → 100 earlier snapshots (250 → 2500 Rg/c): disk structure and jet funnel persist even as the exact turbulent realization diverges — the statistical fidelity a boundary-condition emulator needs.

Same model, different physical variables — drag each slider
Neural operator density forecast
H-AMR density ground truth
H-AMRForecast
◂ ▸

Density ρ — xz slice, r ≤ 50 Rg. Accretion disk + funnel reproduced; errors concentrate in turbulent jet, not the disk body.

Neural operator internal energy forecast
H-AMR internal energy ground truth
H-AMRForecast
◂ ▸

Internal energy ug — same slice. Shock-aware gradient + spectral losses keep disk thermodynamics sharp.

Neural operator magnetic energy forecast
H-AMR magnetic energy ground truth
H-AMRForecast
◂ ▸

Magnetic energy B² — jet/disk boundary preserved with soft ∇·B penalty, no post-hoc cleaning.

Autoregressive rollout movie (1–200 steps). Ground truth vs. neural operator forecast for log10 ρ out to ±200 Rg over 5,000 Rg/c.

03 / Trajectory

Experience & Education

EDUCATION

  • 2025 - 2028 (expected)
    PhD - Physics

    Northwestern University (advisor: Alexander Tchekhovskoy)

  • 2023 - 2025
    MS Computational Science and Engineering; MS Physics

    Georgia Institute of Technology (advisor: Matthew Liska, PhD transferred)

  • 2018 - 2022
    B.Tech - Engineering Physics (Minor in Space and Atmospheric Sciences)

    Delhi Technological University


EXPERIENCE

  • Jun'25 - Present
    Neural-operator emulator for H-AMR GRMHD (thesis)

    Advisor: Alexander Tchekhovskoy, Northwestern University

    Designed a physics-informed GRMHD neural operator with a 9-objective physics loss that forecasts upto 40x faster than simulations.

  • Sep'23 - May'25
    Physics-informed ML for H-AMR

    Advisor: Matthew Liska, Georgia Institute of Technology

    Prototyped physics-informed reconstruction models and physical statistics validity criteria

  • Jul'22 - Jun'23
    Data Engineer (big-data and automated datalakes)

    Genpact

    Built production Snowflake, SQL, and PySpark data lakes and shipped a full-stack legacy-DB-to-Snowflake migrator that cut query runtimes by 17×.

  • May'22 - Aug'23
    GRB Optical and X-ray Plateau Properties Classifier Using Unsupervised Machine Learning

    Advisor: Maria Dainotti, National Astronomical Observatory of Japan

    Clustered Swift-BAT and BATSE gamma-ray burst catalogs using Gaussian Mixture Models and Bayesian Information Criterion across 10+ classes (MNRAS 525, 5204).

  • Nov'21 - Dec'22
    Peering into the Radio Universe - DU IoE project

    Advisor: T. R. Seshadri, University of Delhi

    Helped establish an INR 800,000 radio astronomy lab that focuses on low-frequency antenna fabrication and receiver electronics.

  • May'21 - Aug'21
    Studying dark matter halos and their properties

    Center for Computational Astrophysics, Flatiron Institute, NY

    Case study of IllustrisTNG dark-matter halo merger trees to probe self-similarity across trajectories.

  • May'20 - Jan'21
    Modeling dust scattering and halos using GALEX data

    Advisor: Jayant Murthy, Indian Institute for Astrophysics

    Modeled UV dust scattering and halos in GALEX survey data to study OB star evolution and cosmic dust effects on star-formation rates.

  • Feb'20 - May'22
    Super-resolution (ML) reconstruction of Isotropic turbulent flows

    Advisor: R.K.Singh, Fluid Mechanics Laboratory, Delhi Technological University, India

    Led comparative deep super-resolution benchmarks for isotropic turbulence (ASME IMECE2021 talk & proceedings; Applied-ML Honorable Mention; DeepAI special mention).

04 / Papers & Code

Publications

IN PREPARATION

  • Venkatesh et al. — Physics-constrained Hybrid (2+1)D U-NO for autoregressive GRMHD forecasting on curved spacetime. In Prep

PEER-REVIEWED

  • Bhardwaj, Dainotti, Venkatesh et al. 2023, MNRAS 525, 5204 — GRB optical/X-ray plateau classifier with unsupervised ML. DOI
  • Venkatesh, Pundir 2022, Res. Notes AAS 6, 255 — PERISTOLE: time-delay plots from gravitational lensing. DOICodeZenodo
  • Venkatesh, Bhatt et al. 2021, ASME IMECE2021-69923 — A comparative study of deep learning techniques for spatio-temporal super-resolution reconstruction of forced isotropic turbulent flows. DOI
05 / Talks & Service

Talks & Conferences

SELECTED TALKS

  • Chicagoland AI+Cosmology Workshop — Fermilab Aug 2026
  • Bridging Scales Workshop, Black Hole Initiative — HarvardMar 2026
  • International Mechanical Engineering Congress & Exposition — ASME Nov 2021

CONFERENCES, POSTERS & WORKSHOPS

  • UChicago & Caltech AI+Science Conference 2026 — UChicago PosterOct 2026
  • OpenSKAI Conference — SKAI, Northwestern PosterSep 2025, Sep 2026
  • IAIFI Summer School & Workshop — Harvard & MIT Lightning TalkAug 2025
  • CAMELS Conference — Flatiron InstituteDec 2024
  • Code/Astro — Northwestern Mentor & TA — Parallel ProgrammingJul 2024
  • PyData GlobalOct 2021
  • SciPy ConferenceJul 2021

SERVICE & TEACHING

  • Mentor — NASA Space Apps (Chicago)Nov 2026
  • Reviewer — PEARC: Systems and Systems SoftwareMay 2025
  • Graduate Teaching Assistant — Northwestern, Georgia Tech2023–2026
Industry Resume (PDF) Academic CV (PDF)
Sachin Venkatesh © 2026. All Rights Reserved — Simulation panels & rollout media: author's H-AMR / U-NO research.