Contact

✉️ amil @ucsb.edu

📍 Santa Barbara, California

💻 GitHub

📄 Google Scholar

🔗 LinkedIn

<aside> <img src="/icons/verified_yellow.svg" alt="/icons/verified_yellow.svg" width="40px" /> PhD Student at UCSB Electrical & Computer Engineering Communication and Signal Processing

My thesis is on building and deploying foundation vision models for scientific applications and the art of post-training foundation models. Current projects are 3D Cell Segmentation and Reinforcement Learning for Post-Training Vision Models. I spent the last two summers at the Allen Institute and I am currently in my final year of my PhD.

</aside>

Education

PhD Electrical & Computer Engineering

Advisor: B.S. Manjunath

September 2022 - August 2026

NSF NRT Bioengineering Fellow

• TA for Graduate Computer Vision (ECE/CS 281)
• TA for Undergrad Computer Vision (ECE/CS 181)
• TA for Undergrad Image Processing (ECE 178)

Academic Tree

Academic Tree

Famous Slides

So I have become somewhat famous for my slides and I thought I would share the PDF versions for everyone. I think I spend way too much time on my slides but, at the same time, do you really want to sit through a boring black and white presentation? The honest truth is that no one will sell your work for you, no one will understand how much work you put in to get your SOTA results, and not too many people will take the time after slide 2 to understand what you actually accomplished. Hence why you need to sell. Why is the problem you solved important? Who benefits? How is it done today and what is new in your approach.

BisQue Workshop 2025

I created these slides for our annual workshop on the cloud-based ML Platform that I rebuilt over five years ago. Many of the pictures in the slides are videos showcasing the capability and features of our platform.

BisQue Workshop on deploying foundation models on self-hosted infrastructure (Kubernetes)

BisQue Workshop on deploying foundation models on self-hosted infrastructure (Kubernetes)

ECE Screening Exam Draft Slides

These are the preliminary slides for my screening exam in the Electrical and Computer Engineering Department here at UCSB. This was a one hour oral exam, questions from the public first then up to an hour closed door with committee. Lots of technical discussion, getting into the weeds of how these methods learn, information flow in the transformer, subspace learning, joint attention math, propose a new method for CLIP but for videos, how would you handle temporal information, and similar questions along these lines.

Attention is all you need, CLIP, ViT

Attention is all you need, CLIP, ViT

SuperPoint and SuperGlue

One of the lectures I gave was on SuperPoint and SuperGlue. Creating these slides were such a pain because the papers were quite dense and distilling the information from high level to deep in the weeds definitely tested my knowledge of the core concepts of computer vision.

SuperPoint and SuperGlue ECE 281 Lecture

SuperPoint and SuperGlue ECE 281 Lecture

Selected Publications