I started with basically zero knowledge about stocks. No trading experience. No financial background. I honestly didn't even know what a stop-loss was.
But I got obsessed with a simple question: How do traders actually decide what to buy?
The answer was obvious once I looked — they use technical analysis. Charts, weird-sounding indicators like RSI and MACD, patterns in candlesticks. So I dove in.
I spent months learning. Built spreadsheets to understand momentum indicators. Watched YouTube tutorials at 2 AM. Traded with fake money on Alpaca's free API until the mechanics clicked.
Then one day it hit me: all of this pattern recognition? AI is actually good at that.
That's when something shifted from "just learning" to "I could actually build something here."
Why Silicon Oracle?
Silicon is what chips are made of — the physical substrate of every computer, every algorithm, every AI model that's ever run. Oracle is the ancient idea of a source that gives you foresight, that interprets signals and speaks with conviction.
Put them together: a silicon-based intelligence that reads the market and tells you what it sees. Not a bot that trades for you. A system that thinks alongside you.
The name felt right. So I built the thing.
The Problem I Actually Ran Into
I looked at what was out there — Robinhood, ThinkorSwim, E-Trade. They all cost $10-50 a month. The UIs felt bloated. And worst of all, everyone shares the same rate limits, so everything slows down when the market gets busy.
As for "AI trading bots"? Most of it felt like marketing nonsense.
But then Google released Gemini 2.0 Flash with real-time Search grounding, and it was fast. Actually fast enough to analyze markets as they move. That's when the idea clicked:
What if users brought their own API keys? Then there's no shared rate limits, no API costs I have to eat, no gatekeeping. What if the AI did the analysis, but the human still made the trades?
That was the actual shift from learning project to real product.
What Silicon-Oracle Actually Does
I wanted to combine technical analysis with AI to give traders useful insights. Not hype, not a bot that trades for you — just a tool that actually helps you think.
Real-time analysis. Charts with multiple timeframes, technical indicators, volume data. Nothing fancy — just the stuff traders actually look at, made clean and fast.
Oracle Score™. This is what I'm actually proud of. I built a scoring system that analyzes 15 different factors:
- Momentum (RSI, MACD)
- Trends (moving averages)
- Volatility (beta, sector momentum)
- Volume patterns
- Price action
But here's the thing — it's not the same for everyone. If you trade day-to-day, it weights short-term signals heavily. If you're buying and holding, it looks at longer trends. Same stock, different score for different traders. That felt right.
Email alerts from Gemini. Every hour, the AI looks at the market using your actual portfolio and sends you an email with:
- Stock picks tailored to your style (BUY/HOLD/SELL plus confidence level)
- What new moves actually mean for YOUR positions
- Market catalysts people are talking about (earnings, FDA stuff, economic data)
- Stocks worth watching
Paper trading. You can practice with fake money using real market data. No risk, learn how strategies actually behave.
Position alerts. If something hits your stop-loss or a technical trigger fires, you get emailed immediately. You're not checking the app — the app checks for you.
Everything runs free. Render's free tier. No monthly subscription. No rate limits.
The trade-off: you bring your own API keys from Finnhub, Alpaca, and Gemini. Takes maybe 10 minutes to set up.
The Stack (If You Care About That Stuff)
I kept it pragmatic. Flask for the backend. PostgreSQL via Supabase's free tier. Tailwind + Alpine.js for the frontend. Gemini 2.0 Flash for the AI. Things that work and don't require constant babysitting.
APScheduler handles the hourly jobs — email alerts, monitoring, that kind of thing. GitHub Actions runs tests and deploys automatically. Render hosts it for free.
For code quality: pytest for testing, black for formatting, ruff for linting, mypy for type checking. Boring stuff. The kind of stuff that keeps code from falling apart six months later.
Why This Matters
Most trading platforms lock you in with monthly fees and shared rate limits. Silicon-Oracle doesn't.
You control your own API keys. You own your data. You get unlimited requests. And it costs nothing.
But beyond that, this was never really about trading. It was about learning. How markets work. How to build full-stack products. How to ship something real.
What I Actually Learned Building This
Architecture matters. The BYOK model seemed simple at first but it took weeks to figure out the right way to implement it. Once it clicked though, it was elegant — users get control, I don't bankrupt myself on API costs, everyone wins.
Testing is scary when you skip it. I'm at 25% coverage and I genuinely lose sleep thinking about the untested 75%. I'm spending the next few weeks writing tests on critical paths because shipping untested code is irresponsible. I get it now.
Documentation actually converts users. A good DEPLOYMENT.md and DATABASE_SETUP.md did more for adoption than any marketing could. People want to trust they can self-host this if they need to.
Open source isn't optional for credibility. Putting this on GitHub forced me to clean up code, write proper docs, follow engineering practices. It also meant people could poke holes in my thinking, which made it better.
What's Actually Next
Getting test coverage to 50%+ this month. Improving the backtesting engine. Exploring what a premium tier might look like for traders who want more advanced tooling — though the core will always be free.
The main goal is simpler though: prove you can build real financial software without locking people into monthly subscriptions.
Try It Out
Live demo: https://silicon-oracle.onrender.com/demo
GitHub: https://github.com/HarshShroff/Silicon-Oracle
It's deployed and running. Sign up, add your API keys, start using it. Takes about 10 minutes end-to-end.
I walked in not knowing a support level from a resistance level. Now people are actually using something I built for real money decisions. That's wild.
I'm curious what clicked for you. What started as just learning and turned into something you actually shipped? Drop it in the comments — I read everything.
Silicon-Oracle is MIT licensed open source. Built with Flask, Gemini 2.0 Flash, and the kind of obsession you get when you care about doing something right.
Top comments (0)