I'm an active retail trader. And for years I've been annoyed by the same problem: every good technical analysis tool is either paywalled, locked behind a brokerage UI, or built for institutions — not for someone like me who just wants to see RSI, moving averages, and Fibonacci levels on one chart without paying $300/month.
So I built my own. Meet StockIQ — a free, open-source S&P 500 technical analysis app deployed on Streamlit Community Cloud. No login. No paywall. No nonsense.
And the twist: I built almost all of it with Claude Code as my pair-programming partner.
What StockIQ Does
The app has 10 pages, each answering a question I actually ask as a trader:
SPY Live — Real-time SPY price + a market snapshot covering SPX, Nasdaq, Dow, Russell 2000, and VIX. It's the first thing I open every morning.
SPY Gap Table — One of my favorite features. It tabulates every daily gap in SPY over the last 30 days and shows live fill status during the session. Gaps tend to fill. This page turns that edge into something you can actually act on.
Stock Analyzer — The core of the app. An interactive Plotly chart for any S&P 500 ticker with MA5/20/50/100/200, weekly MA200, RSI, Fibonacci retracement levels, seven auto-detected candlestick reversal patterns, Golden/Death Cross markers, and a composite signal score.
AI Forecast — Claude-powered 10-day directional read on SPY. The app feeds recent multi-indicator context into Claude via the Anthropic SDK and asks for a reasoned forecast. I frame it as a "second opinion," not a crystal ball — but it's genuinely useful to have all the indicators synthesized into one narrative.
Weekly/Monthly Screener — Ranks the top 50 S&P 500 names by candlestick pattern strength.
Bounce Radar — Stocks currently within 5% of their 200-day moving average. A classic mean-reversion setup filter.
Squeeze Scanner — Combines high RSI with elevated short interest. High-risk, high-reward setups.
Strong Buy / Strong Sell — Driven by analyst consensus data.
Munger Watchlist — Quality companies with strong ROE, healthy margins, and low debt trading near their 200-week MA. The long-term compounder filter.
The Stack
-
yfinance+pandasfor market data -
plotlyfor all charts -
anthropicSDK for the AI forecast page - Streamlit Community Cloud for deployment (free tier, zero issues)
Deliberately simple. No database. No auth layer. No infrastructure to maintain. It ships as a Python app and Streamlit handles everything else.
Building with Claude Code: What Actually Happened
This is the part I want to be honest about, because the experience was different from what I expected going in.
What Claude Code is genuinely great at
Indicator math. RSI calculations, Fibonacci level generation, moving average logic, candlestick pattern detection — all of this is well-understood, formulaic code. Claude Code handles it fast and correctly. I didn't have to think about the implementation details at all. I just described what I wanted and reviewed the output.
Boilerplate and wiring. Every Streamlit app has the same skeleton: page config, sidebar, session state, caching decorators, layout columns. Claude Code writes all of this instantly. That alone saved hours across a 10-page app.
Debugging with context. When something broke, I could paste the error plus the relevant code and get a diagnosis in seconds. Not always right on the first try — but usually right on the second.
Where I had to stay in the driver's seat
Architecture decisions. How to structure a 10-page Streamlit app so it doesn't turn into spaghetti. Which pages belong together. What belongs in a shared utility module vs a page file. Claude Code will give you an answer here, but it's not always your answer. I had to push back and redirect several times.
The UX calls. What information density is right for a trader who's glancing at this during market hours? What's the right visual hierarchy on the Stock Analyzer page? These are judgment calls, and Claude Code can prototype options — but it can't tell you which one is correct for your user.
Signal interpretation. The composite signal score on the Stock Analyzer page required me to make actual decisions: how much weight does RSI get vs. MA position vs. pattern strength? Claude Code couldn't answer that. That's trading knowledge, not engineering knowledge.
Data layer gotchas. yfinance has some subtle behaviors that trip up anyone who hasn't worked with it — rate limits, ticker delisting edge cases, timezone handling in intraday data. Claude Code knew the common patterns but wasn't always aware of the specific gotcha I was hitting. I had to investigate some of these myself.
The mental model that stuck
By the end of the project, I'd landed on a framing that felt right:
I own the architecture and the taste. Claude Code accelerates the typing.
That's it. If you go into an AI-assisted build thinking Claude Code will make all the decisions, you'll get a generic app. If you go in with strong opinions about what you're building and why — and use Claude Code to execute those opinions faster — you'll ship something that actually reflects your thinking.
A Few Things I'd Do Differently
Start with a proper page structure. I refactored the multi-page layout twice before landing on the right approach. A bit more upfront planning would have saved that.
Cache more aggressively from day one. Streamlit's @st.cache_data decorator is your best friend for a data-heavy app like this. I added caching incrementally, but I should have built it in from the start.
Build the AI page last. I was excited about the Claude forecast feature and built it early. But the prompt engineering for a useful directional forecast took iteration, and it would have been easier to iterate once the rest of the app was stable.
Try It, Fork It
The app is live at stpicker.streamlit.app — completely free, no account needed. The full source is on GitHub if you want to fork it, run it locally, or adapt it for a different market or asset class.
If you're a retail trader who's been cobbling together multiple tools to get a complete picture — I built this for you. And if you're a developer curious about what it actually feels like to ship a real app with an AI coding agent, I hope the honest account above is useful.
Happy trading. 📈
Built with Python, Streamlit, and Claude Code. Open source, forever free.
Top comments (0)