Salman Farsi

I am interested in the hardest version of every problem I touch.

Not AI in the abstract, but AI that has to make decisions in physical environments, where the cost of being wrong is not a lower accuracy score, but a machine going somewhere it should not. Not software that works in a demo, but software that holds together under real users, real data, and the edge cases nobody thought to write a test for. Not intelligence at scale, but intelligence under constraint, running on hardware that costs almost nothing, in conditions where you cannot simply call a server.

I study Computer Science and Engineering at BRAC University in Dhaka, and my work spans software architecture and system design, machine learning and autonomous systems, robotics, and embedded computing. I move across these areas not because I lack focus, but because the problems I care about exist precisely at their boundaries. Building something that genuinely works requires understanding the entire stack: from the model to the hardware it runs on, from the algorithm to the system around it, and from the backend to the interface a real person ultimately has to use.

What I Work On

My current focus is reliable intelligent systems, particularly where AI has to reason, act, and operate under real-world constraints.

What holds my attention in AI is not simply capability. Models can already reason, perceive, generate, and plan at levels that would have seemed unreachable a few years ago. The harder question is reliability.

How do you build an AI system that behaves consistently, fails gracefully, and can be verified before it does something consequential? How do you keep reasoning and physical action aligned inside an autonomous pipeline? How do you design systems where one model can meaningfully audit another?

These are as much engineering questions as research questions, and they remain difficult problems in real-world AI. That gap is where I choose to work.

Autonomous Systems & Robotics

Autonomous systems interest me because they remove the comfortable assumption that a human is always in the loop.

A language model that produces a slightly wrong answer is inconvenient. An autonomous system that produces a slightly wrong action in a physical environment is a fundamentally different problem.

Working in that space requires thinking across perception, reasoning, planning, control, and feedback rather than treating them as isolated components. My work in autonomous driving, Vision-Language-Action systems, reinforcement learning, and robotics comes from that interest: understanding how intelligent systems can move from perception and reasoning to actions that are actually consistent, executable, and safe.

Software & Systems Engineering

On the software engineering side, I think about architecture the way a structural engineer thinks about load.

Systems that feel clean when you build them and stay clean as they grow are not accidents. They are the result of deliberate decisions about how data flows, how components communicate, how failures propagate, and how the system behaves when reality diverges from the original design.

Reliability is not a feature you add at the end. It is a property you either design in from the beginning or spend the rest of the project trying to recover.

That mindset influences how I approach everything from software architecture and backend systems to AI pipelines and embedded applications.

Beyond Engineering

Outside engineering, I shoot photography, travel whenever I can, and hike.

Being somewhere unfamiliar and moving through it slowly—whether with a camera or on a trail—trains a kind of attention that I find valuable in engineering as well. You notice what changes, what does not, what you initially overlooked, and what only becomes visible when you stop moving too quickly.

I think good engineering requires much the same thing:

curiosity, patience, and the willingness to look closely at what everyone else has overlooked.