⭕️ About
I’m currently a computer science MMath student at University of Waterloo and fortunate to be advised by Prof. Lukasz Golab. I have been collaborating closely with Prof. Mohammad Mohammadi Amiri. I recieved my bachelor degree in Computer Science at Sharif University of Technology. I was a scientific intern at Max Planck Institute for Software Systems, Germany where I had the great pleasure to work with Professor Laurent Bindschaedler on Graph anomaly detection and blockchain.
📌 Research Interests
My main areas of interests include:
- Large Language Models
- AI Safety and Alignment
- Data Centric AI
To get a comprehensive understanding of my Research and Work experiences, as well as my Projects, please take a look at my detailed descriptions.
Besides my professional pursuits, I have a deep passion for reading. I consider myself a true bookworm and find joy in immersing myself in books, expanding my knowledge, and exploring different worlds through literature. Reading is not only a hobby for me, but also a way to relax, learn, and broaden my perspective. You can view some of my favorite books in the Misc. tab.
In addition to books, I also enjoy listening to podcasts.
🗞 News
[10/2025] Our paper titled A Median Perspective on Unlabeled Data for Out-of-Distribution Detection is now available on arXiv.
[10/2025] Gave a talk titled “We Are Losing the War for Real Data, It Is Time to Mark the Originals” at SoLaR @ COLM 2025 (October 2025).
[08/2025] Our paper titled The Alignment Game: The Inevitable Conflict of Values in Generative Models was accepted by Conference on Language Modeling (COLM) Workshop on Socially Responsible Language Modelling Research (SoLaR), Montreal, Canada, 2025.
[07/2025] Our paper titled SentenceKV: Efficient LLM Inference via Sentence-Level Semantic KV Caching was accepted by Conference on Language Modeling (COLM) Montreal, Canada, 2025 for publication! code
[07/2025] Received the José Blakeley Scholarship in Data Systems ($10,000) from University of Waterloo.
[06/2025] Our paper titled Embrace the diversity: Avoiding mode collapse with polarized curation in generative retraining was accepted by International Conference on Machine Learning (ICML) Workshop on DataWorld: Unifying data curation frameworks across domains, Vancouver, Canada, 2025.
[06/2025] Our paper titled Fragile by Design: Formalizing Watermarking Tradeoffs via Paraphrasing was accepted by International Conference on Machine Learning (ICML) Workshop on Technical AI Governance, Vancouver, Canada, 2025.
[01/2025] Started my Master’s at the Cheriton School of Computer Science, University of Waterloo!
Random fact: I haven’t used paper for reading and writing for the past 6 years. I’m all about digital note-taking and ebooks, you know? 📚💻
Let’s do some calculations. If we assume that every student goes to class for 180 days and takes an average of 6 classes, with 2 pages used for note-taking, brainstorming, and other activities, and 2 pages used for assignments and homework for each class, we can figure out that each student uses about 3000 pages per year. Considering that each school has 200 students, and based on online research suggesting a total of around 5 million schools worldwide by 2023, schools are effectively using 3 trillion pages each year. Roughly speaking, each tree produces 10,000 sheets. So, if we calculate it, schools around the world are using 300 million trees each year, assuming no recycling happens. Math doesn’t lie though…