Interactive slides that execute Python in your browser, an AI tutor that answers questions about the page you are reading, and live six-week cohorts that turn self-study into a finished credential.
An approach Prof. Xuhu Wan has been refining since 2012, when he first introduced live, runnable code in a Year-1 Business Statistics course. Every course on this site is self-developed and offered to the public. Students registered in his university course can access the matching materials here.
Dynamic incentives, information design, and game-theoretic markets — applied to trading, disclosure, contracts, and multi-agent AI.
The first principles of how LLMs reason — chain-of-thought, self-consistency, ToT / ReAct / Reflexion, process reward models, and the new RL-trained reasoning models (o-series, DeepSeek R1, Gemini Thinking). Practitioner's reference: build, evaluate, and deploy reasoning systems. Capstone: train a tiny reasoning model end-to-end.
Collaboration, incentives, and markets when the system has more than one AI in it. Game theory, MARL (CTDE / MAPPO / QMIX), self-play, social dilemmas, LLM-agent frameworks (AutoGen / CrewAI / LangGraph), MCP & A2A protocols, mechanism design for AI agents, agent markets. Capstone: negotiating marketplace of LLM agents.
An MIT PhD-level course in information, mechanism, and computational market design — full proofs, problem sets, and primary-source reading lists. Persuasion & concavification, Bayes-correlated equilibrium, robust / prior-free MD, dynamic mechanisms & contracts, matching, auctions & platforms, data markets & privacy, financial information design, algorithmic & computational market design, data-driven MD, and AI-agent alignment.
Mechanism design, strategic LLMs, and autonomous markets. What happens when the "agent" in principal-agent / market design / auctions / bargaining is an LLM or RL system? Each chapter: formal model, theorem-level results, diagrams, two frontier-paper deep-dives, and an open research direction.
A frontier PhD course on L2 AutoResearch — AI-executed, human-verified economics. How strong economists use AI to compress the research cycle while designing incentive-compatible verification against hallucination, fake citations, weak novelty, and non-reproducibility. Each chapter: an AI workflow, a human verification task, an in-class lab, and a paper idea. Capstone: a mini AI-augmented economics paper with full provenance log.
From the trader between the ears to live execution — reasoning & edge, portfolio construction, model risk, microstructure, derivatives, deep learning for quant, and the new-data & crypto labs.
Strict backtesting and machine learning for portfolio management, grounded in López de Prado's Advances in Financial Machine Learning and Machine Learning for Asset Managers. Information-driven bars, triple-barrier labeling & meta-labeling, sample uniqueness & sequential bootstrap, fractionally differentiated features, MDI/MDA/clustered feature importance, purged k-fold & combinatorial purged CV, covariance denoising, and the heart of it — evaluating the Sharpe ratio honestly (Probabilistic & Deflated Sharpe Ratio, multiple-testing correction) and measuring backtest overfitting (CSCV / PBO), then robust portfolio construction (HRP / NCO) and a strict end-to-end protocol. Every result is computed live in the page. Self-study, from first principles.
From formula search to self-improving research agents — a genealogy of automated alpha discovery. Genetic programming (AlphaEvolve), RL (AlphaGen), generative–predictive mining with dynamic combination (AlphaForge), GFlowNets (AlphaSAGE), and the new LLM research loops (RD-Agent(Q), AlphaAgent), with each paper's architecture and core equations rendered live and the AlphaForge codebase digested. Closes on the methodological spine the literature skips — Deflated Sharpe, PBO, and multiple-testing haircuts — and a proposal: an agent that searches over prediction tasks, conditioned on events, with selection-bias control wired into the loop.
The trader between the ears. Fast EV (Fermi for traders), calibration as a trainable skill, Bayes in practice, Kelly from first principles, fractional Kelly & drawdown, sizing under uncertainty about your own edge, prop-bet practice, IC as the language of edge, decision quality vs outcome (the Resulting trap), and the one-page research memo. Capstone: a multi-week paper book scored on calibration · growth rate · memo quality — the three numbers a senior market-maker actually uses.
A manager-level course on turning forecasts, trades, and mandates into a capital allocation process. Mandate & strategy universe, alpha-to-capital, risk budgeting across sleeves, portfolio constraints, covariance under stress, position sizing, factor control, liquidity / capacity / impact, turnover & rebalancing, drawdown management, robust optimization (MVO / risk parity / HRP / Black–Litterman), multi-strategy allocation, PnL & risk attribution, stress testing, governance, live dashboard, and IC communication. Capstone: a manager dashboard and allocation engine that scores each sleeve scale / hold / cut / kill.
The statistics of not fooling yourself. Information Coefficient diagnostics, drift detection and retraining, multiple-testing corrections (Bonferroni, BH, Deflated Sharpe), and the methodology discipline used by top hedge funds.
How electronic markets actually work, and how to design execution algorithms that survive spreads, impact, queues, and latency. Limit order books, VWAP/TWAP/POV, Almgren–Chriss optimal execution, market making (Avellaneda–Stoikov), event-driven backtesting, and live TCA — capstone: a full LOB simulator comparing execution strategies under different liquidity regimes.
Practical derivatives for quants, market makers, and multi-asset researchers. No-arbitrage pricing and put–call parity, Black–Scholes with the Greeks and hedging, the implied-volatility surface (smile, skew, term structure), volatility forecasting (GARCH/HAR/IV vs RV), options market making, exotics and VIX products. Capstone: an options analytics dashboard — price options, fit a vol surface, and backtest the variance risk premium.
Modern deep learning for noisy financial prediction, from tabular alpha models to sequence models and order-book forecasting. MLPs, LSTMs / TCNs / Transformers, DeepLOB, cross-sectional ranking, self-supervised pretraining, reinforcement learning for execution, and leakage-safe validation (purged CV, embargo, deflated Sharpe). Capstone: train a deep model and turn it into a cost-aware, drift-monitored trading strategy.
Find, collect, clean, and test non-traditional data sources for tradable signals. Prediction markets (Polymarket / Kalshi), event calendars, news/text feeds, web & social attention, satellite / AIS / foot traffic, entity mapping, point-in-time engineering, event studies, IC and capacity, and kill rules. Capstone: build a new-data signal and write a hedge-fund-style research memo.
Build real-time crypto trading signals from exchange microstructure and on-chain flow. Crypto market structure (spot / perps / funding / liquidations), Hyperliquid WebSocket ingestion, order-book reconstruction, aggressor flow, funding & basis carry, open interest and liquidation pressure, on-chain wallet & stablecoin flows, DEX/LP signals, CEX↔DEX arbitrage, live monitoring and alerts. Capstone: a live monitor combining Hyperliquid order flow with on-chain wallet signals.
Where code meets people and things — generative design, interaction, and computation that reaches into the physical world.
The orientation course for the whole track. For anxious, talented Year-1 / Year-2 Industrial Design students: where the field is going, which careers are growing, which are vulnerable to AI automation, which skills to build over the next three years, and how to prepare for internships and employment. Evidence-based, balanced realism — not doom, not hype. With curated videos, designer interviews, and a final personal AI-era career strategy.
Creative coding for the visual arts — drawing, color, motion, and generative design in the browser. Build sketches and interactive visuals from first principles, one runnable example at a time.
Computational Interaction & Intelligent Products — for industrial & interaction designers. Build interactive objects, responsive environments, and AI-enhanced products: the sense–think–respond loop, generative form, physical computing (Arduino, prototyped in p5), spatial interaction, AI systems, and calm/ambient interfaces.
Interactive systems, interfaces, product ecosystems, motion, and scalable design workflows — inside Figma. Built for Pratt Industrial Design and adjacent fields (UX, healthcare, smart products, creative tech, AI products, interaction design): far beyond mobile UI into dashboards, devices, multi-device ecosystems, motion as communication, and design systems that scale across teams. Deep prose teaching, runnable p5 simulations of Figma concepts (auto layout, variants, motion), curated YouTube videos, real-Figma studio briefs.
AI Creative Technology for designers — code as a creative material. Build generative systems, behavioral motion, camera/audio interaction, AI media, and intelligent products through live coding cells and creative experiments.
A computational studio for building WORLDS — procedural environments, explorable & cinematic space, autonomous agents and flocking, living ecosystems, and AI that directs the experience. The spatial, systems-level companion to AI Creative Coding. Builds on Creative Coding 1 & 2.
Professional digital product workflow for Pratt-style Industrial Design students. Computational form and CAD, parametric & generative geometry, material storytelling, physical-digital interaction, fabrication-aware design, intelligent product systems, and AI-augmented design futures — taught as design reasoning, not button-clicking. Rhino, Grasshopper, Blender, Figma, Arduino, Cura, KeyShot referenced naturally throughout.
Game design for industrial designers — from object interaction to playable systems. Learn experience goals, game feel as interaction feedback, shape language, systems thinking, and playtesting; treat code, AI, and Godot as prototyping materials. Play a real game inside the lesson, edit live GDScript and press Run to change it, and finish a portfolio-ready playable concept package. No coding required to start.
The maker's hands. AI art & sound pipelines, real GDScript in Godot, and a production process — culminating in a finished, original entry to a live game jam. The project-based sequel to Game Design I, for Pratt Industrial Design students. (Outline live; full lectures in development.)
The full BA track — from intro to advanced, domain modelling, and alternative data (networks, text, social, modern AI).
From DataFrames to decision models. Covers the Python data-science stack (pandas, NumPy, SciPy, Matplotlib), DataFrames in depth, linear regression with CAPM and Fama-French factors, and clustering for customer segmentation. Worked examples use real market data.
The working analyst's toolkit, from pandas to alpha. Series and DataFrames in depth, markets as data objects (returns, vol, drawdown), simple and multi-factor regression, Fama-French, regularization, gradient boosting, walk-forward backtesting. Based on the book Modern Business Analytics.
Distributions and EVT, predictive models, causal inference (RCTs, IV, RDD, DiD), Bayesian methods + MCMC, time series with ARIMA / GARCH / cointegration, clustering, pattern recognition, embeddings, foundation models, interpretability and fairness.
Schemas → data models → ontologies, in pure Python (NetworkX, rdflib, Neo4j, pgvector). Entity resolution, bitemporal modelling, GraphRAG, GNNs on knowledge graphs, EU AI Act compliance.
Network analysis (nodes, edges, centrality, PageRank, community detection, network formation) plus text analytics (topic models, sentiment, cascades). All browser-executable.
LLMs, foundation models, multimodal analysis, dynamic networks, knowledge graphs, agents, vector DBs, recommenders, MLOps, misinformation, deepfake detection, financial systemic risk.
A hands-on introduction to general Python — no pandas, no NumPy. Variables, control flow, functions, collections, files. Every chapter ends with a coding practice (not a multiple-choice quiz) that runs in your browser.
For working programmers who already know basic Python. Object-oriented design at depth (descriptors, metaclasses, ABCs, Protocols), a structural/algebraic view of class morphisms, and how Python talks to the internet — sockets, HTTP, async I/O. Coding practice in every chapter.
An introduction to statistics — exploratory data analysis, probability, random variables, sampling distributions and the Central Limit Theorem. Visual intuition and animation over code.
A continuation. Confidence intervals, hypothesis testing for means and proportions, and simple linear regression — fitting, assumptions, R², and inference on slope and intercept.
Linear algebra as a reference. Vectors, matrices, OLS via least squares, norms & cosine similarity, projections, eigenvectors and PageRank, SVD and PCA, network & covariance matrices. Every concept paired with the numpy one-liner that computes it — open any chapter, find the formula and the code.
Each chapter is structured prose with code cells interleaved. Click Run on any cell to execute it in your browser; edit numbers, re-run, and watch the chart change.
Predict-then-reveal cells force active recall. Debug-yourself exercises have hidden tests that pass only when your fix is correct. Spaced-repetition cards bring back what you've forgotten.
The AI tutor lives on every page. It knows the chapter you're on, the code you just ran, and the errors you just saw — so its answers are specific to your context, not generic. And the slides themselves update with your questions — over a cohort, the deck grows the explanations the cohort actually asked for.
Eumathe Academy is the personal teaching channel of Prof. Xuhu Wan. Every course is developed independently — designed from scratch for self-directed learners, expanded with the historical and economic context that textbooks usually omit, and made fully interactive so you can learn at your own pace.