Agent systems
Memory, tools, delegation, and observable workflows for reliable action in the world.
AI researcher · Systems builder · Author
Wayland Zhang
I research adaptive intelligence: how AI can compute dynamically, learn continuously, preserve personal memory, and reason beyond a single context window.
Alongside this research, I build systems in two domains where mistakes have real consequences.
Quantitative systems
Mathematical and neural models, point-in-time evidence, execution-aware testing, and independent risk control for decisions in markets.
The methods may connect; the claims must be proven separately.
My writing and teaching have reached more than 100,000 practitioners and led to three technical books and two open-source platforms.
Contact: waylandzhang[at]gmail.com
Current Work
Agent Systems
Reliable delegated action across local and cloud agents: tools, permissions, replay, and observable workflows.
Memory & Reasoning
Dynamic computation, continual learning, and personal memory for intelligence that must persist beyond a single context window.
Quantitative Research
Adaptive multi-mechanism signals, inverse-problem framing, execution-aware validation, and independently governed risk.
Mathematics and neural networks can supply shared questions and tools. Results do not transfer across domains; every market claim must earn its own point-in-time, cost, and forward evidence.
Background
Previously founded a social advertising platform acquired by AdChina (Alibaba Group). Studied Computer Science at the University of Toronto ('03–'07), where I attended Professor Geoffrey Hinton's class.
Systems I Build
Production-grade multi-agent platform with deterministic replay, budget enforcement, and enterprise observability. Built with Rust, Go, and Python.
AI agent runtime powered by Shannon: local tools, memory, permissions, and complex task delegation through observable workflows.
AI-native quantitative research and trading lab for adaptive multi-mechanism signals, point-in-time evidence, execution-aware validation, and independently governed risk.
Selected Research & Writing
All essaysMathematical foundations
Chern's Question, Answered by Two Worlds at Once
How piecewise-smooth manifolds and overlapping covers connect rock mechanics, neural networks, and the mathematics of continuous and discontinuous worlds.
Neural-network foundations
The Four Realms of Neural Networks
From PDE solvers to manifold geometry and measurement: a framework for thinking about learned representations and personal intelligence.
Quantitative research
The Geometry of Alpha
Why the durable quant edge is a factory that keeps producing and validating signals, not a single strategy recipe.
Agent systems
The AI Agent Harness
What production agent runtimes converge on when context, tools, permissions, verification, and the loop become the product.
Quantitative systems
The Backtest-to-Live Gap
How to attribute the gap across signal, data, execution, fees, financing, and market regime before changing a strategy.
GitHub Activity
Books
Timeline
2026.07~Building Dnalyaw, an AI-native quantitative research and trading lab, while researching AI inference and training memory models.
2026~Building Kocoro and Shannon for reliable delegated AI work, and writing about production agent engineering.
2025~Researching dynamic neural networks, continual learning, and reasoning beyond fixed Transformer loops.
2023–2024Became deeply interested in large language models and general AI.
2019–2022Took a break from startups while actively investing in several challenging projects.
2015–2018Founded a cloud-native platform designed to simplify cloud infrastructure management.
2008–2014Founded a social advertising platform later acquired by AdChina, a subsidiary of Alibaba Group.
2003–2007Studied Computer Science at the University of Toronto and attended Geoffrey Hinton's class.