Sep 2021 — Jul 2025
Education
BS Computer Science
Lahore University of Management Sciences
I am pursuing an MS in Computer Science and Engineering at the University of Michigan, Ann Arbor, after completing my BS in Computer Science at the Lahore University of Management Sciences (LUMS) in 2025. From 2025–26, I also served as a Teaching Fellow (junior faculty) in the LUMS CS department.
Current focus Making compact reasoning models more capable, interpretable, and dependable.
Path so far
Sep 2023 — May 2024
Research
Research Intern
CSaLT, LUMS
Jun 2024 — Jul 2026
Research
Research Lead
CSaLT, LUMS
Jan 2025 — Jul 2026
Research
Research Lead
City @ LUMS
Jul 2025
Milestone
Completed BS Computer Science
Lahore University of Management Sciences
Jul 2025 — Jul 2026
Teaching
Teaching Fellow
LUMS · Junior Faculty
Sep 2026 — Present
Education
MS Computer Science & Engineering
University of Michigan, Ann Arbor
Research
01
Mechanistic distillation
Leading a joint CSaLT and City @ LUMS effort to distill reasoning from Qwen-3.5-4B into Qwen-3.5-2B. Our student currently outperforms base knowledge distillation on ARC-C, GSM8K, MMLU, MATH-500, and HellaSwag.
02
Code evaluation
Building a code benchmark that evaluates LLMs through novel games—testing whether models can reason, implement, and adapt beyond familiar problem sets.
03
Reward hacking in RLVR
Searching for causal explanations of reward hacking during reinforcement learning with verifiable rewards.
Teaching
CS-200 · Fundamentals of Object-Oriented Programming
Co-taught 400 first-year students; designed and managed 14 C/C++ labs across two sections, three programming assignments, lecture tutorials, and assessments including two midterms and a final.
CS-202 · Data Structures
Co-taught 300 sophomores in C++, delivered weekly lecture tutorials, and designed and graded two midterms and the final exam.
MS AI Bootcamp
Taught 100 incoming graduate students from diverse backgrounds: Python and NumPy, probability foundations, and a multivariable-calculus refresher.
CS-5306 · Generative AI
Lead TA for a graduate course, revamping its curriculum and programming assignments: LSTMs for word embeddings, a SmolLM built from scratch, and PEFT with LoRA and QLoRA.
CS-436 · Machine Learning
Teaching assistant for the undergraduate machine learning course.
Interests
Distillation
Isolating select teacher capabilities and amplifying them in smaller student models.
Mechanistic interpretability
Exploring weight-based interpretability beyond SAEs, inspired by Goodfire’s work on interpreting LM parameters.
RLVR
Tracing causal links between an agent’s behaviour and reward hacking in reinforcement learning with verifiable rewards.
