ground truth photo
ground truth rendered gaussians
WebGL unavailable — rendering the rasterized fallback (the content still works!)
3DGS · training live
iteration0
loss (L1 + λ·D-SSIM)
gaussians0
lr (position)1.6e-4
SH degree0
PSNR·SSIM·LPIPS
cameraview 000/167
← lossinitialization
σ scale1.0×
initializing 9,216 gaussians from SfM noise…

SEEMANDHARJAIN

PhD student @ UC San Diego, advised by Prof. Manmohan Chandraker. I build multi-agent LLM systems that do 3D vision research autonomously — and this page is rendered the way I see the world: as a cloud of gaussians, optimizing itself into something meaningful. Watch it train. →

scroll = next stage · drag = novel view · drag ⇆ = GT compare
stage 01 / initialization

Point cloud → person

Every 3DGS scene starts as structureless noise; so does every research career. Mine initialized at IIT Indore (B.Tech, CSE), where I founded HealthAIgnite, a healthcare computer-vision consultancy — my first densification step.

The optimizer then routed me through an MS at UIUC (GPA 3.92/4) with Prof. David Forsyth — single-shot 3D reconstruction, convex decomposition, intrinsic images, relighting with diffusion models — and a Siebel Scholar (Class of 2024) award.

Today I'm a PhD student at UC San Diego working at the intersection of LLM agents and 3D vision: multi-agent systems that read papers, write code, run experiments, and discover new methods on their own. Recent splats in my covariance: Nerfify (CVPR 2026 Highlight) which turns NeRF papers into executable code, Splatify — its 3DGS counterpart with knowledge discovery — and Idea2Paper, a pipeline that goes from literature to novel method to validated paper, end-to-end.

Along the way: Google (Student Researcher, BigML — optimized SANA diffusion for Veo, built Gemma-2 captioning), Aireal (Research Scientist, CV), DOCOMO Innovations, and Salesforce (AMTS, AI).

stage 02 / optimization

What my gradients flow toward

🤖

Autonomous Research Agents

Multi-agent LLM frameworks that reproduce, discover, and innovate: papers → trainable code → new methods. Nerfify, Splatify, Idea2Paper.

Radiance Fields & 3DGS

NeRFs, 3D Gaussian Splatting, single-shot reconstruction, convex decomposition — explicit and implicit 3D, and the tooling to automate it.

Generative 3D & Diffusion

Diffusion models for 3D-aware image/video synthesis, relighting, intrinsic decomposition; faster sampling with better geometric consistency.

stage 03 / densification

Converged publications

// each paper below is one cluster in the galaxy behind this text — 12 clusters, 12 papers

stage 04 / adaptive density control

Recent densification events

Apr 2026Nerfify selected as a CVPR 2026 Highlight! 🎉
Mar 2026Invited talk at Google on NERFIFY.
Jan 2026Invited talk at Pixel Cafe.
Jun 2025Student Researcher (BigML) at Google, New York — optimized SANA diffusion for Veo, built Gemma-2-based captioning.
Jul 2024Started PhD at UC San Diego with Prof. Manmohan Chandraker.
May 2024Joined Aireal as Research Scientist (CV) — single-image-to-3D textured mesh, AI interior design (Livvy).
Sep 2023Selected as Siebel Scholar, Class of 2024.
stage 05 / the optimization trajectory

Loss curve of a career

2024 → now

PhD, Computer Science · UC San Diego

Advisor: Prof. Manmohan Chandraker. LLM agents × 3D vision: autonomous research systems for NeRF/3DGS, generative 3D.

2025

Student Researcher · Google (BigML, NYC)

Optimized SANA diffusion for Veo; Gemma-2-based captioning at scale.

2024

Research Scientist (CV) · Aireal

Single-image → textured 3D mesh; AI interior design product (Livvy).

2022 → 2024

MS, Computer Science · UIUC

Advisor: Prof. David Forsyth. GPA 3.92/4. Siebel Scholar '24. 3D reconstruction, convex decomposition, relighting.

2021 — 2022

Salesforce (AMTS, AI) · DOCOMO Innovations (Research Intern)

Applied ML in production; large-scale contact tracing & recommendation research.

2017 → 2021

B.Tech, CSE · IIT Indore — founded HealthAIgnite

Healthcare-AI consulting focused on computer vision. Iteration 0.

stage 06 / novel view synthesis

Render a new view — together

The splats behind this card just spelled it out: I'm always up for collaborations on agents × 3D, radiance fields, and autonomous science.

sejain [at] ucsd [dot] edu

d densify · p prune · r re-train · 1–4 render mode · 2×click local split · drag = novel view