About me

Hi, my name is Mudit Dhawan, and I am pursuing my Master’s in Machine Learning (MSML) at Carnegie Mellon University (CMU). My current research spans neuroscience and machine learning with Prof. Michael J. Tarr, I am analyzing category representation in pre-trained neural models and the human brain, aiming to understand representational alignment across artificial and biological systems.

Most recently, I was a Machine Learning Engineer Intern at Adobe, where I developed SLM-based conversational agents for tool orchestration using supervised fine-tuning and GRPO, achieving 95%+ reduction in time-to-market for production-ready agents. I also built a synthetic data generation pipeline leveraging LLMs (GPT, LLaMA 70B) to simulate diverse, grounded user interactions. Our main goal was to study fine-tuning methods under resource constraint settings, along the axes of model size, and number of training data points.

Previously, under the supervision of Prof. Ruslan Salakhutdinov, I work on controllable generation in diffusion models, focusing on text-to-music diffusion and its vulnerabilities to jailbreaking, researching on memorization in diffusion models. We explored stable-audio-2, and AudioLDM models along with textual inversion to produce copyrighted music.

Prior to my graduate studies at CMU, I was a Research Fellow at Microsoft Research India (MSRI) advised by Dr. Manik Varma, working on efficient and scalable machine learning for real-world deployment—including a query-autocomplete reformulation deployed on Bing AI Chat and Search and a low-latency recommendation algorithm that outperformed production baselines. We introduced an efficient cross-encoder based ranking methodology which was successfully filled as a US Patent.

I completed my B.Tech. in Electronics and Communication Engineering at IIIT Delhi, where I was a Research Assistant at Precog advised by Prof. Ponnurangam Kumaraguru (“PK”), working on AI for Social Good with a focus on multimodal misinformation detection.

If you are interested in research collaboration in neuroscience-inspired ML, generative models, or retrieval systems, please email me.