I built a project just to learn Rust: a 3D animated trading floor. Thanks to help from AI it went further than I expected.
PitCrowd is an open-source trading floor where the crowd in each pit moves with the price. Here’s how a Rust newcomer got it built and why it’s now yours to take further.

PitCrowd animated trading floor
Before screens took over, markets were rooms. Traders wearing colourful jackets stood in octagonal pits, shouting and signalling with their hands, and you could see a market: when a price broke, the crowd surged; when it stalled, people drifted off. It was called open outcry. My initial idea was to see if visualising the market like this could provide some kind of advantage over traditional charts.
I also wanted to learn Rust, and to find out what Jev, an AI model from TypeSafe that reads market conditions, could do. So I gave myself a project far bigger than my experience level should have allowed for: rebuild that room, and wire it to live prices.
The project grew and grew and after starting to understand some key Rust fundamentals I asked Claude to step in and help me take it to the next level.
The result is PitCrowd. I think it is a good base to build on and I’ve taken it as far as I can on my own so I am open sourcing and writing this article. I’d love to see how others expand on it.
What it is
The floor has nine pits, one per market: gold, silver, the Nasdaq, the S&P 500, the Dow, the FTSE, the euro, sterling and Bitcoin.
- The crowd follows the price. A sharp move pulls traders into the pit and it gets loud. A quiet market empties.
- Paper traders walk the floor. Each strategy is a figure that walks to the pit it’s trading, with its record on a leaderboard.
- You can trade too. A trade ticket lets you place paper trades alongside them.
- Bitcoin’s order book stands in the pit. The buy and sell orders waiting in the market appear as walls of liquidity, and as a heatmap panel.
How a newcomer built it
I’ll be upfront: I’m not an experienced Rust engineer. I built PitCrowd with an AI coding assistant as my pair. I set the direction and made the decisions; it wrote a lot of the code; and I had to understand, question and test everything it suggested before it went in.
Working that way on something this size (about 54,000 lines in the end) taught me more than anything else I have tried to learn Rust. What made it work:
- Tests as guardrails. Around 350 of them. When code is cheap to write, the thing that matters is knowing whether it’s right. Every feature came with checks, and CI runs them on Linux and macOS on every change.
- Small steps I could follow. Asking for one change at a time, and asking why when something surprised me, kept me in charge of the design instead of just accepting whatever arrived.
- Rules that don’t bend. Some things were never up for negotiation: the API key never goes into the browser, nothing places a real order, and nothing in the tests costs money. Writing those down early meant every new piece got checked against them.
- Being willing to throw things away. An early version of the built-in test market accidentally had a pattern in it that some strategies could exploit. It was caught, cut, and a test now stops it creeping back.
What I learned about Rust along the way
I didn’t learn Rust from a book first. I learned it because of what I was building, and wrote it down as I went: the repo includes 25 short lessons, from ownership and borrowing through traits, async and channels, to how a game engine organises a world and how to work with an AI model like Jev from Rust.
- The compiler is a teacher. The borrow checker felt like a wall at first and like a reviewer after that. Whole classes of bug just don’t compile.
- One codebase, two targets. The floor is built with Bevy, a Rust game engine, and the same code runs as a desktop app or in a browser tab via WebAssembly.
- Data first. Bevy uses an entity-component-system design: instead of objects that own their behaviour, you write systems that run over data. A pit’s crowd, a trader’s walk and a price board are all just queries over the same world.
How it fits together
The tape is the floor’s only source of prices: one small server with one simple protocol. The relay exists so an API key never ships inside a web page. And the evolving pit generates trading strategies by the thousand, backtests them on years of price history, and seats the survivors on the floor as paper traders.
Real data, and running without it
My own floor ran on a live feed from a cTrader demo account. The floor doesn’t care where prices come from: anything that can make an HTTP request can drive a pit. The repo has an adapter that streams real Bitcoin, and its order book, from Kraken’s free public API, plus a template and a walkthrough for wiring in your own source.
When I prepared the public version, I also wanted it to run for anyone, without an account or a price feed. So I added a synthetic market: it makes up realistic-looking prices for all nine pits, with busy and quiet hours, sudden jumps and an order book.
I made sure it had no hidden patterns: the direction of every move is pure chance. That started as a practical choice, but it turned out to be useful. Anything that “makes money” on it did so by luck, so it shows you what luck looks like. There’s plenty of room to develop it, whether that’s making it behave more like a real market (panic selling, realistic reactions to news) or deliberately hiding a pattern in it, to see whether the strategies can find one.
Honest about money
Nothing here makes money, and I’m not claiming it does. PitCrowd trades on paper only.
When I ran 39 paper strategies on real prices, 10 of them were in profit. That sounds promising until you think about coin flips: get 100 people to flip a coin ten times, and a few will get eight or nine heads with no skill at all. Ten winners out of 39 was no more than luck would give you.
That’s why the evolving pit works the way it does. The more strategies it tries, the higher the bar a strategy has to clear, and only its record going forward, not its backtest, can earn it a permanent place. I wrote those rules down before its first real run, so I couldn’t move the goalposts once results came in.
And as for my original question, whether seeing a market as a crowd gives you an edge over a chart: I found that I if I spent a fair amount of time on one market watching the comings and goings of the traders I could start to get a small feel for that market. Maybe that is the same as looking at a chart, but this is a far more intuitive and fun way to watch a market. Maybe with more data being fed in and more market mechanics it could develop into something that could help traders get more of a feel, like the tape readers of the past. Of course the question is still open for anyone who wants to explore it.
The part I’m least sure about: Jev
Jev classifies market conditions from the context you give it. I’m new to it, and I doubt I used it as well as it can be used. My experiments didn’t find an edge, but I’d put that down to my inputs and inexperience before I’d put it on the model. Someone who knows Jev well could do far more here.
Jev is optional and paid. Without a key, the floor uses a simple free rule in its place and labels it as such.
Where it could go
I’ve taken it as far as I can alone. The roadmap is a menu, not a promise:
- Configurable markets: today the nine pits are fixed. Making the list configurable would let any feed bring its own instruments.
- A browser demo: run the synthetic market inside the page, so anyone can try it from a link.
- A better synthetic market: more realistic behaviour, or hidden patterns to test the strategies against.
- Use Jev better: better inputs, or Jev as a filter on a strategy that already works.
- More adapters: OANDA, Interactive Brokers, Binance, Alpaca.
There are good first issues, a contributing guide, and Discussions are open. If you build on it, a link back is appreciated.
PitCrowd on GitHub : https://github.com/Bwizrd/PitCrowd
Paul Leppard is a senior software engineer who builds things outside his day job to learn. PitCrowd is a research and visualisation project, not trading software, and nothing in it is financial advice.



