school CS Undergrad @ Stony Brook University

Hi, I'm
Adam Jacuch.

Moving beyond Transformers, I engineer novel architectures with the goal of solving agentic intelligence.

See my work Read philosophy arrow_forward

							class
							Agent(nn.Module):

							  def
							forward(self, x):

							    # O(N) Complexity

							    return
							self.ssm_layer(x)
						

speed 3.97 GPA

memory PyTorch + CUDA

4.0 Core GPA

JAX Enthusiast

ICSC Finalist

About Me

Lead. Founder.
Architect.

I am a Computer Science undergraduate at Stony Brook University, but my work extends far beyond the classroom. I engineer novel AI systems with the ultimate goal of solving intelligence.

My focus is on architecture design—decoding how structural choices dictate capabilities, and combining them to create unprecedented models. Read more.

PyTorch Core CUDA Systems C++ Robotics

precision_manufacturing

Vex Robotics Team

Lead Programmer

Developing autonomous routines and control systems for competitive robotics.

groups

CRIZM VIP Team

Junior Founder

Helped established a vertically integrated project team focused on novel AI applications.

school

Stony Brook University

CS Undergraduate

Specializing in AI Systems and Low-Level Optimization. 3.97 GPA.

Agentic Intelligence

Beyond Transformers

First Principles

psychology

Autonomy over Prediction

Developing systems that don't just predict the next token, but reason, plan, and execute multi-step workflows autonomously.

bolt

Breaking the O(N²) Barrier

Investigating sub-quadratic architectures to achieve linear scaling in sequence length without sacrificing recall capabilities.

build

Built from Scratch

I believe in building libraries (like Axiom) from zero to truly master the mathematical foundations of gradient descent and backpropagation.

Selected Research

Engineering at the intersection of theory and architecture.

Research Paper

Beyond Parameter Scaling

Introducing recursive depth: a parameter-efficient scaling axis that uses shared recurrent blocks to enable precise associative routing. Boosts MQAR accuracy substantially and improves Mamba-2 perplexity on FineWeb-Edu without adding parameters.

Coming Soon arrow_right_alt

deployed_code

Axiom

A named-axis tensor calculus library for JAX. Compiles into efficient JAX code.

View Repo arrow_right_alt

tune

Optimizer Playground

Interactive web app for visualizing loss landscapes and experimenting with various network optimizers in real-time.

Try App arrow_right_alt