Tariq Shams
Senior Software and Artificial Intelligence Engineer
Creator of TAUI (Terminal Agent UI) Standards
▶TAUI Details
I am building TAUI (Terminal Agent UI), an open-source standard for autonomous agents in the terminal. Most agents today interact with shells by directly writing raw text or ANSI escape codes. It’s brittle and prone to hallucinations.
Generating executable UI code is dangerous (injection vectors) and token-expensive. I believe we should move from "text buffers" to "structured state documents".
The TAUI Ecosystem
Core standard for Grid, Panel, and Table primitives.
Zod-powered engine for safe agent-TUI compliance.
Unified state and event runtime for terminal agents.
React bridge rendering high-fidelity TUI sessions.
I bridge the gap between latest research and product development. I build and own full stack applications backed by robust AI systems and scalable infrastructure.
I spend my free time reading textbooks, research papers and technical blogs on deep learning, reinforcement learning, agentic patterns, and physics. I love building core technologies from scratch to deeply understand their inner workings.
Awards
First Place, Vibe Hack Hackathon
NVIDIA GTC 2026
Achieved first place at the Vibe Hack Hackathon, competing against Ivy League students and engineers from Big Tech. Received an autographed RTX 5080 by NVIDIA CEO Jensen Huang.
Watch this livestream with NVIDIA where we discuss winning the hackathon, our model architecture, and present a live demo: Watch here.
Blog
Giving a fly's wiring a game to play
A walkthrough of Google's role in mapping a fruit fly's nervous system, and how I used that wiring to build a browser experiment with Space Invaders, Tetris, and a controller learning from scratch.
Score-Coupled Narrative Poisoning
A proposal for why autonomous research systems should separate trust in a verified artifact from trust in the explanation attached to it.
Determinism in Transformer-Based Language Models
Temperature zero seems deterministic. Floating point arithmetic and dynamic batching explain why real inference systems can still disagree.
Large Language Models vs their Harness
Large Language Models are a useful tool that can help us with several tasks. But your llm isn't writing to your filesystem. We need to build harnesses around them to make them useful.
Notes and reflections from NVIDIA GTC 2026
Reflecting on my time at NVIDIA GTC 2026, winning the Vibe Hack Hackathon, and deep dives into Agentic AI, Inference Engineering, and more.
Building a Image Editing Pipeline
Building a Image Editing Pipeline with Qwen Image-Edit and Qwen-Image-Layered
So you want to learn AI
A guide to make the transition from Software Engineering to AI Engineering
When efficient code fails: Timing Attacks
Efficient code is normally what we professional software engineers strive for, but sometimes it could make us vulnerable to timing attacks.
Recursion - Part I - Beginners
A technique that is often neglected, Recursion can be a beautiful way to solve certain classes of problems.