ALPHA

AI-native hedge fund for real-time, news-aware, data-driven trading.

Comment

Inspiration

In the 1980s, a small group of funds started using computers to analyze markets. It seemed unconventional at the time, but it birthed the massive quantitative trading industry we know today. We are at a similar inflection point with AI. During my time as a quant researcher, I saw firsthand that the world's largest funds are too slow to adapt; when I asked compliance to let us use ChatGPT, I didn't even get a response.

The future isn't just bolting AI onto existing strategies. It is building entirely new funds driven by autonomous agents to discover new sources of alpha. For the Finance Track, I wanted to build a system that moves beyond basic analysis to actively synthesize market trends and predict movements using an AI swarm. However, raw data is often overwhelming. For the Figma Track, my goal was to take massive, intimidating financial datasets—like 10-Ks and real-time price histories—and transform them into a highly interactive, aesthetic story that feels intentional and usable, rather than just a dashboard of disconnected charts.

What it does

ALPHA is a 100% client-side web application that operates as your personal AI hedge fund team. There is no login, no account requirement, and no database.

The user experience follows a seamless, story-driven flow:

The Hook: Users land on a clean interface featuring a neural network animation, immediately establishing the AI-native branding.

Ticker Selection: Users can type in any custom ticker symbol or select a "Quick Pick" (e.g., NVDA, AAPL).

Data Ingestion: The system automatically fetches Yahoo Finance data in the background. Users have the option to upload SEC filings (10-K, 10-Q) or earnings call transcripts to provide the AI with deeper context.

Agent Swarm: Clicking "Deploy Agent Swarm" triggers a processing animation while the client-side analysis engine runs its evaluations.

The Dashboard: The user is presented with a comprehensive 4-tab view (Overview, Signals, Agents, News) that tells the complete story of the asset. It features a master "AI Conviction Signal," interactive price charts across seven timeframes, and targeted insights from six distinct AI agents.

Compare Mode: Users can input a second ticker to view a side-by-side Risk Matrix and Financials comparison.

How we built it

To ensure a beautiful, story-driven user experience, I mapped out the entire visual hierarchy and data flow in Figma Make before writing any code.

Technically, ALPHA is a entirely serverless application. All state is handled in-memory.

Frontend Architecture: Built with React 18, Vite 6, and Tailwind CSS v4.

UI & Visualizations: I utilized Recharts for price, revenue, and EPS data, Radix UI (shadcn/ui) for accessible components, and Motion for fluid interface animations.

The Data Engine: The application relies on a highly robust, multi-tier API fallback architecture. It attempts to pull live data from Yahoo Finance v8 as the primary source. If that fails, it cascades to Financial Modeling Prep, then Twelve Data, and finally falls back to rich, hardcoded fundamental data and synthetic price histories. This guarantees the dashboard always populates.

Challenges we ran into

Building a complex financial analysis tool entirely on the client side presented severe technical roadblocks:

Strict CORS Policies: Yahoo Finance heavily blocks browser-origin requests. To resolve this, I engineered a 3-proxy fallback chain (corsproxy.io → allorigins.win → codetabs.com). Because these public proxies frequently drop, I had to implement individual timeout controllers for each to ensure application stability.

Client-Side PDF Parsing: Without a backend to process uploaded SEC documents, I built a custom, zero-dependency PDF text extractor directly in TypeScript. It parses raw bytes, handles FlateDecode stream decompression, and extracts text from BT/ET operators entirely in the browser, bypassing the need for heavy external libraries like pdf.js.

API Rate Limits & Complex State: Free API tiers rate-limit aggressively. This forced me to build a complex state-threading system without relying on a state management library. The yahooFundamentals object had to flow cleanly from the document upload component into the analysis engine, successfully merging live API data on top of hardcoded baselines so the UI never displays null values.

Accomplishments that we're proud of

I am incredibly proud to have built a complex, multi-agent AI system completely on the client side as a solo developer in a single weekend. Successfully bypassing strict CORS limitations and building a raw PDF byte parser from scratch were major technical wins.

From a product design perspective, I successfully translated my Figma Make concepts into a working application. I took highly complex quantitative data and built an interface that is visually stunning, highly interactive, and tells a clear, decisive story about a stock's potential.

What we learned

This hackathon pushed my technical limits in frontend architecture and network management.

I gained a deep understanding of browser CORS policies and how to architect resilient proxy-chain fallbacks for unreliable APIs.

I learned how to manipulate raw byte streams and decompress files directly within the browser using native APIs.

I significantly improved my ability to manage complex, cascading application state using pure React hooks (useState, useEffect, useCallback) to route data seamlessly across components.

What's next for ALPHA

The immediate next step is to transition beyond in-memory state. I plan to add persistent database storage so users can track their portfolios and saved analyses over time. Following that, I intend to integrate live paper trading APIs (like Alpaca) so ALPHA's "Conviction Signals" can be automatically executed and backtested against live markets. Ultimately, I want to expand the AI's data ingestion to include alternative data streams, giving the client-side agents an even sharper informational edge.

Built With

Share this project:

Updates