Turn client workflows, risk, controls, and failure scenarios into operable systems.
Public evidence · No accounts, positions, or private infrastructure
OPERATIONS / SYSTEMS / AI-AUGMENTED DELIVERY
Isaac Cheng
I turn complex financial operations into clear systems, testable prototypes, and useful tools.
I use operational judgement and technical connectivity experience to frame the problem, then AI agents to accelerate research, prototyping, testing, and delivery. Each case states my role and its limits.
Define clear boundaries, states, and acceptance evidence across business, operations, and technology.
Use agents to expand research and implementation speed while keeping judgement, testing, and accountability human-owned.
Evidence reel / 25 seconds
Five capabilities. Evidence first.
Each chapter comes from a separate real project and keeps its evidence maturity visible.
Trading systems in progress
Not a collection of disconnected crypto demos, but a capability chain from market observation and research to risk-controlled execution.
01 / ObserveSimpleTerminal
Knowledge layer / Trading Technologies Knowledge BaseFIX · Exchange Connectivity · Market StructureUnify quotes, depth, trades, and derivatives state across five venues.
02 / ResearchStablecoin Risk MLTest market assumptions with machine learning, event studies, and tail risk.
03 / ControlExecution & Risk LayerVenue-agnostic paper validation, deployment checks, risk rules, and diagnosable state.
04 / ExecuteVenue Adapter + Live EngineMove from one adapter contract to confirmed-state execution; UMX now, major venues on the roadmap.
Selected work
Four distinct forms of evidence, separating professional work, independent products, academic research, and concept demos.
01
Multi-venue algo architecture and risk systems
Multi-Venue Algo Trading Architecture
A venue-agnostic algo trading operating layer connecting adapter contracts, research, backtesting, paper and controlled live execution, risk, and observability, with UMX as the current first implementation.
EvidenceVenue adapter contract · Backtest → Paper → Controlled Live · Observable controls
Explore public architecture →02
Real-time cross-market engineering
SimpleTerminal — Multi-Exchange Market Terminal
A zero-backend, zero-API-key market workspace that brings live quotes, depth, trades, and derivatives context from five exchanges into one screen.
Evidence5 exchanges · 0 API keys · 1 unified workspace
Open live product ↗03
Academic research and risk interpretation
Machine Learning for Stablecoin Depeg Risk
An eight-year study comparing Logistic Regression, Random Forest, LSTM, SHAP, and three VaR methods across different stablecoin architectures.
EvidenceEight years of data · Out-of-time validation · Honest model limits
Open live product ↗04
Operational workflow and full-stack rebuild
Cha Chaan Teng Operations System
How an ordering exercise became an operations system, connecting mobile guest ordering and server-side pricing to a staff Kanban and traceable order state.
EvidenceGuest order → Server pricing → Staff Kanban → State history
V0 · FLASKOrdering form
V1 · LIVE DEMOOperations flow
From evidence to a next step
Hiring teams and collaborators need different information. Both paths begin with public evidence and keep sensitive details private.
Exploration map
Different subjects, one recurring question: how can complex information become a system people can understand and use?
My publishing loop
Daily learning stays private first. It becomes public only after editing, visualisation, and a sensitivity review.
01Private notes
02Safety review
03Visualise
04Publish
Every new item is private by default. Content reaches the website only after a human review explicitly clears it for publication.
Recent build logs
What I built, what broke, what I learned, and what I would change next.
Hermes Agent Teams: Why Role Isolation Clicked for Me
Instead of asking one AI to research, judge, write, and build at once, give each agent a clear role, isolated context, and explicit handoff.
How This Site Works: Obsidian, AI Agents, and Safe Publishing
Daily Markdown stays in private Obsidian first, then passes editing, sensitivity review, and visualisation before it reaches the public site.
Learning in progress
MSc Data Science2024–2026 · Distinction
AI agent systemsOrchestration, verification, content workflows
Product and interface designFrom prototype to useful tool
Data visualisationMarket data, models, and interpretation