STARFISH — AI-Powered Stock Analysis

Inspiration

"In this AI era, knowledge has no value — only the people who know how to push the right buttons at the right time succeed."

That single idea built STARFISH.

The rich get richer. The poor keep getting poorer. That is not an accident — it is a system. For centuries, the tools of wealth creation were locked behind expensive degrees, exclusive networks, and institutions that only let you in if you already belonged. The game was rigged before you sat down.

But every era has exactly one window — one moment where the rules have not yet been rewritten and the walls have not yet been rebuilt. This is that window. AI has ripped the gate off its hinges. The analyst in the hedge fund and the kid with a laptop and a dream now have access to the same intelligence. Not a watered-down version. The same.

I am not building STARFISH from a comfortable position of privilege with safety nets and connections. I am building it because this is the only era in human history where someone like me — with no capital, no credentials, no network — can compete at the highest level and win. The window is open right now. It will not stay open forever. The moment institutions figure out how to re-close it, they will.

So I am moving fast. And I am building something that does not just help me rise — it hands the ladder to everyone else standing where I stood.

I am not a trained developer. I am not a certified financial analyst. Six months ago, I could not write a single line of Python. But I watched the markets every day, felt the frustration of expensive analysts, paywalled research, and tools built only for people who already knew everything — and I asked a different question: what if I just didn't need to know?

You do not need to know how a bicycle works to ride one. You just need to know where to put your feet. AI is the bicycle. STARFISH is the seat. I built an entire production-grade financial analysis platform — real-time charts, multi-model AI reasoning, live news feeds, technical indicators — without a computer science degree, without a finance degree, without years of experience. I pushed the right buttons. The AI took care of the rest.

And here is the important truth that this project proves: AI will only get smarter. The gap between "expert" and "curious person with the right tools" is closing to zero. STARFISH is proof of that collapse happening in real time.

What It Does

STARFISH is a live stock intelligence platform that puts institutional-grade analysis in the hands of anyone with a ticker symbol.

Core Features:

Interactive Charting Engine — Candlestick and line charts with real-time Yahoo Finance data. Overlay SMA (20/50/200), Bollinger Bands, RSI, MACD, and Volume indicators with a single click. Fully responsive, zero-latency rendering with Plotly.
Multi-Model AI Analysis — Powered by OpenRouter, users can run deep trading analysis through three distinct AI engines: DeepSeek R1 (chain-of-thought reasoning), Llama 3.3 70B (balanced speed), and Qwen3 Coder (quantitative focus). Each model receives a structured 30-day OHLCV feed, computed indicator values, trend signals, and macro context — and returns a structured verdict: BUY / SELL / HOLD, with entry price, stop loss, two profit targets, technical breakdown, news & macro context, risk factors, and a step-by-step trader action plan.
Live Rate Limit Dashboard — Real-time RPM and RPD tracking per model, with animated progress bars so users always know which AI engine is available.
Live Financial News — Embedded YouTube live streams from CNBC TV18, Bloomberg Global, and Yahoo Finance, switchable in one click.
Popular Stock Chips — One-click loading for AAPL, GOOGL, MSFT, TSLA, NVDA, TCS.NS, Reliance.NS, and more.
Global Market Support — Indian NSE/BSE stocks supported natively via .NS / .BO suffixes alongside all US markets.

How I Built It

The stack is deliberately lean and powerful:

Layer	Technology
Backend	Python / Flask
Data	Yahoo Finance (custom scraper with crumb auth + yfinance fallback)
Charts	Plotly (server-side render → HTML div injection)
AI	OpenRouter API (DeepSeek R1, Llama 3.3, Qwen3)
Indicators	NumPy / Pandas (SMA, EMA, BB, RSI, MACD, ATR — all computed from scratch)
Frontend	Vanilla HTML/CSS/JS — zero frontend framework, zero build step
News	YouTube embed API with live/latest detection
Rate Limiting	In-memory sliding window (per-model RPM + RPD tracking with threading locks)

The AI prompt engineering was the deepest technical work. Each analysis request packages 30 days of OHLCV data, all computed indicator values, trend signal arrays, 52-week high/low distances, volume ratios, ATR volatility, and Bollinger Band %B — all formatted as a structured multi-section prompt that forces the model to return a single valid JSON object with exactly defined keys. No markdown, no preamble, no hallucinated structure.

The Yahoo Finance scraper uses a three-layer fallback: v8 chart API → v7 CSV download → yfinance library, with session/crumb caching, random user-agent rotation, and 30-minute cache TTL to handle Yahoo's aggressive bot detection.

Challenges I Ran Into

1. Yahoo Finance Anti-Bot Walls Yahoo Finance actively blocks scraping. Building a robust three-layer fallback with crumb authentication, cookie management, randomized user agents, and graceful degradation took more iterations than any other part of the project.

2. Getting AI to Return Clean JSON Every Time Large language models want to talk. Getting them to return only a valid JSON object — no markdown fences, no preamble, no trailing commentary — required careful prompt engineering, post-processing regex stripping, and error-handling for malformed outputs.

3. Rate Limit UX Three AI models, each with RPM and RPD constraints, visible to a user who shouldn't need to understand API rate limiting. Building the animated progress-bar dashboard that makes this feel intuitive (not technical) was a real design challenge.

4. Building Without Prior Knowledge The most honest challenge: I started this not knowing Python, Flask, or financial APIs. Every single line came from understanding what I wanted to exist, articulating it clearly, and iterating. This is the challenge I am most proud of overcoming.

Accomplishments That I'm Proud Of

Shipped a production-grade financial platform with zero prior coding experience. This is the headline. It is real. It works. It is deployed.
The AI analysis quality is genuinely useful — not toy output. The structured prompts extract real multi-paragraph technical breakdowns with specific price targets, risk factors, and action plans that reflect actual market context.
The design is clean enough to compete with paid tools. Dark, minimal, typographically precise — STARFISH does not look like a hackathon project.
Proved the thesis live. STARFISH is not just a project about AI democratizing expertise — it is the demonstration. Built by someone with no background, it does what expensive Bloomberg terminal add-ons do.

What I Learned

I learned that the bottleneck is no longer technical skill — it is clarity of vision.

The person who can describe exactly what they want, in precise enough terms, with enough domain understanding to validate the output — that person can build anything. The AI fills the implementation gap. What I brought was the idea, the judgment, the taste, and the persistence to iterate until it was right.

I also learned:

How financial data APIs work and why they are fragile
How technical indicators (RSI, MACD, Bollinger Bands, ATR) are actually computed — not just what they mean, but the math
How to structure prompts for structured JSON output from LLMs
How rate limiting, session management, and API authentication work at a systems level
That shipping something imperfect and real is worth more than planning something perfect that doesn't exist