🤖📈 Building an AI Trading System with Claude Code — A Weekend Experiment That Taught Me a Lot
A personal story about what's possible when curiosity meets today's AI tools — building **TradeAIQ**, an experimental AI trading system, on weekends with Claude Code, free market data, and open-source Python libraries. And an invitation for anyone with an idea to give it a try.
A personal story about what’s possible when curiosity meets today’s AI tools — and an invitation for anyone with an idea to give it a try.
| Investment | Value |
|---|---|
| Time | Weekends only |
| Hosting Cost | $0 |
| Out-of-Pocket | 1 domain name |
Table of Contents
- Why I’m Sharing This
- The First Real Hurdle: Token Limits
- The System: TradeAIQ
- How the Intelligence Flows
- Working with Claude Like a Team
- What It Cost
- A Peek at the Result
- What I Took Away
1. Why I’m Sharing This
Like many of us, I’ve carried product ideas around for years. In the past, turning an idea into working software usually meant needing a team, a budget, and a lot of time — things most of us with day jobs and families simply don’t have.
What’s changed recently isn’t me. It’s the tools.
Agentic AI tools like Claude Code have lowered the barrier so much that small teams — and even individuals working on weekends — can now explore ideas that once felt out of reach. I wanted to test that for myself with an honest experiment, and I’m sharing the journey in case it encourages someone else sitting on an idea to take the first step.
💡 Why a trading system? Not because I’m an expert — precisely because it’s a genuinely complex problem: live market data, mathematical rigor, machine learning, an AI reasoning layer, and a public-facing UI. If the approach worked here, it could work for many kinds of ideas.
2. The First Real Hurdle: Token Limits
If you use a personal AI tool on a basic subscription, you know this pain: token limits. Long, sprawling sessions burn through your quota quickly, and a complex system simply can’t be built in one giant conversation.
The workaround that saved me was nothing revolutionary — just good old software engineering discipline: modular, contract-driven development.
- 🧭 Design first — I sketched the overall system myself (my thought process, the ideas, the structure), then shared it with Claude Code to refine and improve together.
- 🔁 Iterate on the architecture — until the design felt right.
- 🧩 Break the system into small, independent modules — each buildable within a single session, each with a clean, stable contract to its neighbors.
This solved the token problem, and it turned out to be a great fit for working with AI agents in general: give them well-bounded modules with clear interfaces, and they do remarkably well. Nothing here is my invention — it’s a pattern anyone can apply.
3. The System: TradeAIQ
An experimental AI trading system with predictors for Swing, Day, and Intraday trading — live at TradeAIQ.app.
Architecture at a Glance
📡 1. Market Data Server
The foundation — and my first lesson in patience. I started with Yahoo Finance, hit data quality issues, and after more searching landed on Alpaca’s free market data, designing carefully around its rate limits.
One design decision I’m glad I made early: the API always returns the same output shape regardless of the backend source — Alpaca, Yahoo, Robinhood, or anything else. With a consistent contract, every downstream module is insulated from data-source changes. If the backend needs to change tomorrow, nothing else breaks.
📐 2. Mathematical Engine
Built on the excellent open-source Python trading and ML libraries the community has created over the years. This module pulls data from the Market Data Server and computes the technical indicators the system needs. I stood on the shoulders of a lot of generous open-source contributors here.
🧠 3. AI Brain
A Claude Code plugin that handles what pure math can’t: edge cases like sentiment analysis and contextual judgment calls that don’t reduce to a formula.
🖥️ 4. UI Layer
A public-facing interface deployed on Cloudflare Pages — chosen because this is an experiment, and free hosting made it possible to share openly without worrying about costs.
4. How the Intelligence Flows
The screener output is a layered pipeline rather than a single model’s guess:
Mathematical Engine (pure technical computation) → ML Model (trained on actual outcomes) → AI Brain (covers edge cases like sentiment analysis) → Published publicly on the hosted system
Reports run periodically and publish automatically.
The internal portal’s Report Schedulers — persona-scoped watchlist decision reports running on their own cadence, each writing a versioned JSON report that the public site then picks up and publishes.
5. Working with Claude Like a Team
One of the most enjoyable parts of this experiment was learning to collaborate with Claude Code in different roles:
- 📋 Business Analyst — helping clarify requirements and priorities
- 🏛️ Architect — keeping the modular design and contracts honest
- 🛠️ Specialist Engineers — backend, frontend, and everything in between
It never felt like replacing anyone. It felt like getting to learn from the collective knowledge of many disciplines at once — and I came out of each session understanding my own system better. If anything, this experience deepened my respect for the engineers, analysts, and architects whose expertise these tools are built on.
6. What It Cost
| Investment | Cost |
|---|---|
| My weekends and personal time | The real investment |
| Claude Pro subscription | Already had it |
| Cloudflare Pages hosting | $0 |
| Domain (TradeAIQ.app) | The one small purchase |
Essentially free, apart from time and a domain name — which is exactly what made it possible to try at all.
7. A Peek at the Result
🔗 See it live: https://tradeaiq.app/app
Here’s the live system — the layered pipeline in action. Every row you see is the output of the Mathematical Engine, ML model, and AI Brain working together: fused-decision ranking, conviction scores, risk/reward ratios, trend sparklines, entry prices, market regime detection, and grades.
🏠 The Public Home
The public face of the experiment — Swing, Day, and Intraday predictors, all published automatically.
📊 The Dashboard
The dashboard rolls up the latest published run: market regime, top-ranked setups, and how the pipeline is currently reading the market.
🎯 The Candidates Screener
The Swing Trader Candidates view: S&P 500 symbols ranked by fused decision, each with strategy setup (S/R Breakout, RSI Divergence, EMA Crossover, and more), conviction, R:R, trend, and regime — with rich filtering across side, grade, market cap, EPS trends, and forecast direction. Refreshed automatically and published publicly.
🔍 The Symbol Detail View
The Symbol Detail view: a full candlestick chart with support/resistance levels, entry/stop/target overlays, RSI/MACD/volume panels — and on the right, the decision panel showing the setup with its conviction score, the checks behind it (HTF alignment, confluence, quality, R:R), the AI scoring layers, price-action confirmation criteria, and a complete trade setup with risk sizing.
8. What I Took Away
This experiment wasn’t about proving I could do something special. It was about discovering that the gap between having an idea and building it has quietly become much smaller — for all of us.
I made plenty of mistakes along the way. I hit walls, backtracked, and redesigned things more than once. But every hurdle had a workaround, and the process itself was deeply satisfying.
If you’re someone with an idea you’ve been putting off because you thought you’d need a team, a budget, or more expertise — I hope this story nudges you to just start:
- 🧭 Design first — think it through before you type
- 🧩 Break things into small modules — one session, one module
- 🔌 Keep your interfaces clean — contracts protect you from change
- 🎁 Lean on free tools and open-source libraries — the community has done a lot of the hard work already
- 🤝 Let AI agents share the load — treat them as teammates, not magic
The best part of this new era isn’t what any one person can build. It’s how many more of us get to build at all.
⚠️ TradeAIQ is an experimental project shared for educational purposes. Nothing here is financial advice — do your own research before making any investment decision.
Built on weekends with Claude Code, Alpaca free market data, open-source Python libraries, and Cloudflare Pages. Live at TradeAIQ.app. If you’re working on something similar, I’d love to hear about it.