Adway Kanhere

Hey there! Welcome to my slice of the web - part dev notebook, part research shenanigans and the rest about coaxing GPUs to behave.

I’m a first-year CS PhD student at The Johns Hopkins University. I previously worked as a Research Engineer at the University of Maryland Institute for Health Computing where I developed ML/AI for translational biomedical applications in surgery, radiology, and healthtech infra.

In the past, I have occasionally lectured as an Adjunct Instructor at JHU’s Carey Business School. Before that, I received a master’s degree in biomedical engineering, also from Hopkins, where I learned the fine art of taking really expensive 3D selfies of internal organs via MRI and CT. Turns out there’s decent interest in professional photography of the human body, as long as you call it “medical imaging.”

Research:

My research interests lie at the intersection of artificial intelligence, neuroscience and medical imaging. Feel free to connect with me if you want to share a thought!

CS Research:

• Multimodal learning, Vision-Language foundational models, and reasoning: Data and compute optimization strategies for LLMs/LMMs using non-convex optimization techniques.
• Surgical world models: Physics-aligned generative models of surgical scenes that capture instrument-tissue dynamics and procedural state, enabling downstream applications in skill assessment, simulation, and pre-operative planning.
• Long video generation: Scaling video diffusion models to minute-scale, temporally coherent generation through test-time training, memory-augmented architectures, and curriculum-based fine-tuning strategies.
• Federated learning: Harmonizing heterogeneous non-i.i.d data in a model agnostic and modality agnostic setting while maintaining SOTA privacy measures.

Clinical Research:

• Biomarker discovery: Identifying novel biomarkers to predict cognitive outcomes based on gross anatomy
• Medical image analysis: Building segmentation models for translational applications in radiology and surgery using foundational models

Entrepreneurial Detour:

I was also the co-founder of Farmplus, a dairy-tech startup where we built a cool fitbit for cows, tracked it with good ol’ telecom frequecies, snagged 1st place at the 2022 Hopkins New Venture Challenge (HOPSTART) and were subsequently incubated at the JHU Spark Incubator. Check out the BTS story in my blog (because my other co-founder and best friend wants me to take public accountability)

IRL:

When the GPUs cool down and my attention span times out, I swap keyboards for hand wraps and go a few rounds with a punching bag, or log suspicious “cardiovascular spikes” at Centennial Park/JHU Homewood campus. Apart from cardio, I also indulge in video and acoustic production sessions where I attempt to merge film-score strings with indian classical strings like the santoor and sitar, and emerge as something my friends describe “formally interesting.” Once a month I also attempt a full digital detox in the mountains. Just me, my playlist, and whatever existential bug reports nature files. It’s amazing how well our brain’s neural networks converge when the only feedback loop is nature telling you to chill.