Amil Agrawal

Hello! I study computer science & math at Brown University.

I’m interested in reinforcement learning, world models, and efficient inference.

In industry and personal work, I've also deeply explored robotics, recommender systems, LLM post-training, and large-scale inference infrastructure. I love meeting new people, so feel free to say hello through email or LinkedIn below!

At Brown

Internships

Snap Inc.

ML Research Engineer Intern

Applied research on L1 recommendation foundation models for 500M+ Snapchat user feeds.

AMD

ML Engineer Intern

Inference infrastructure optimization and kernel-agent post-training for ROCm.

RTX

Robotics Engineer Intern

Multi-agent reinforcement learning for collaborative aerial autonomy.

Bloomfilter

Software Engineer Intern

First SWE intern, working on AI and core product systems.

RSGbot

A Minecraft speedrunning agent.
Screenshots in, keyboard and mouse actions out.

Run notes

Excerpts from longer agent runs, with thinking intervals removed or accelerated. The strips replay timestamped agent commands: highlights show explicit requested controls, while higher-level skills remain named actions. These archives do not contain measured key-down or mouse-event traces. Select a clip for its source and outcome.

RapidTinyRec

Optimizing MiniOneRec inference with shared KV caches, decode replay, and folded attention heads.

Original → optimized FP32

868.8→114.8ms median

7.6× median speedup. Same rankings; separate recorded runs.

Original loop · FP32Optimized FP32
Loading recorded requests…

Original → experimental BF16

868.8→56.5ms median

15.4× median speedup. Rankings change; separate recorded runs.

Original loop · FP32Experimental BF16
Loading recorded requests…

Same 500-user cohort · recorded measurements.

Other comparisons & measurement notes

Overall comparisons align the original released-loop timings with later optimized timings by cohort index. Same Office checkpoint, 500 users, RTX A6000, batch 1 and beam 50; measured in separate runs, not a new paired experiment. Speedups are ratios of the two measured medians, not multiplied per-step gains. The optimized FP32 run matches all 500 original ranking lists; scores are not bitwise identical. BF16 is experimental and changes rankings. Model loading, compilation, capture and warmup are excluded. Traces show 80-request windows in cohort order and retain all measurements. The individual optimization comparisons keep their own paired controls; the isolated BF16 comparison marks its 693.8 ms outlier above the 90 ms axis. Original timings · FP32 evidence · Precision & quality

SysVRC

Autonomous VRC routines in ROS 2 and Gazebo. The field, controller estimate, and command trace on one timeline.

Gazebo skills recording

On the field

6121C · Spin Up

Autonomous skills
Driver skills

Different Field Setup Used

PAVER

Panoramic Alignment and View-based Evidence Recovery

Recovering occluded street-view pixels from neighboring panoramas, using feature matching and geometry.

Before: the capture car obscures road markings in Strasbourg After: partial road recovery using observed pixels from neighboring panoramas
Strasbourg · capture-car removal
Before: a controlled foreground mask beside a Beijing boulevard, with buildings and the road visible around it After: saved donor-view recovery of the masked foreground in the same synthetic Beijing boulevard scene
Beijing · controlled occlusion
Quai Saint-Nicolas

A · find the visible road pixels

Image sources & method

Strasbourg: Eurométropole de Strasbourg / Panoramax, Etalab Open Licence 2.0. Partial capture-car removal with DISK + LightGlue. Unsupported pixels remain unchanged.

Beijing: Yijing Guo et al., PanoCity / PanoVGGT, CC BY 4.0. A controlled mask in a synthetic scene; saved tool-assisted donor registration and color calibration, separate from the automatic pipeline.

Map geometry · ODbL 1.0. Beijing is reprojected from the saved 1024 × 512 working panorama; its limited detail and residual seam are preserved.

Map © OpenStreetMap

Reconstructing Mercury

Reconstructing Mercury’s southern gravity field with lunar transfer learning and topography.

Input · degree 25

Gravity · mGal
Gravity color scale: blue for negative, white for zero, red for positive
−2000+200
Mercury’s gravity field at spherical harmonic degree 25

Southern detail is a model prediction.

wild-env

A wildfire-response environment for multi-agent learning: fire spread, aircraft dispatch, suppression, and logistics.

Viewer & simulation notes

The application’s own Qt interface, playing a recorded synthetic foothill scenario at four simulated minutes per second. The demonstration changes vehicle selection and suppression layers using the viewer’s original controls. Greedy assignment baseline, seed 20260726; this is not a trained policy or a real-fire forecast.

stemp

Four stems, twelve pads, a step sequencer.
A music instrument for macOS.