Skip to content

PRD.md

Latest commit

 

History

History
259 lines (208 loc) · 9.58 KB

PRD.md

File metadata and controls

259 lines (208 loc) · 9.58 KB
Raw

AlphaForge — Product Requirements Document

Vision

AlphaForge is a quant trading platform where anyone can build, backtest, and compare trading strategies using traditional algorithms, LLMs, or a hybrid of both — all without risking real money until they're ready.

Prove it works. Then deploy it.

Problem

  1. Algo trading tools are intimidating — QuantConnect requires C#/Python expertise, most platforms assume you're already a quant
  2. AI trading tools are reckless — platforms like Cod3x let you deploy AI agents with real money and no backtesting
  3. No platform lets you A/B test AI vs traditional strategies — everyone picks a side, nobody compares properly
  4. The gap: an accessible platform that combines visual strategy building, AI-powered strategies, rigorous backtesting, and a safe simulation-to-live pipeline

Target Users

Phase 1: Builders & Tinkerers

  • Developers curious about algorithmic trading
  • Crypto/stock traders who want to automate but don't trust black boxes
  • Data science students exploring quantitative finance

Phase 2: Broader Retail

  • Non-technical traders who want AI-assisted strategies via visual builder
  • Communities/groups who want to compare and share strategies

Core Features

1. Strategy Engine

Three modes of strategy creation:

🔢 Algorithmic Strategies

  • Built-in library: RSI, MACD, Bollinger Bands, mean reversion, momentum, pairs trading
  • Combine indicators with logical rules (AND/OR/threshold)
  • Custom Python scripts for advanced users

🧠 LLM Strategies

  • Feed market data (OHLCV, volume, indicators) + optional news/sentiment to an LLM
  • Model selection: Claude, GPT, Gemini, local models
  • Prompt template system — define what the LLM sees and how it responds
  • Structured output: BUY / SELL / HOLD with confidence + reasoning

🔀 Hybrid Strategies

  • Algo generates candidate signals
  • LLM acts as filter/confirmation layer
  • Or: LLM proposes, algo validates risk parameters
  • Configurable pipeline: define which component does what

2. Visual Strategy Builder

  • Drag-and-drop node-based editor
  • Nodes: data sources, indicators, conditions, LLM blocks, actions
  • Connect nodes to build strategy flow
  • Preview output at each node
  • Export to Python for advanced customization
  • No code required for basic strategies

3. Backtesting Engine

  • Run strategies against historical data
  • Configurable parameters:
    • Date range
    • Starting capital
    • Position sizing rules
    • Commission/fee model (per broker)
    • Slippage model (realistic fill simulation)
  • Walk-forward testing (train on period A, test on period B)
  • Monte Carlo simulation for robustness testing
  • Overfitting detection warnings

4. Paper Trading Simulator

  • Real-time simulation using live market data
  • Identical interface to live trading
  • Virtual portfolio with realistic execution
  • No money at risk
  • Track performance over days/weeks before going live

5. Live Trading (Phase 2)

  • Broker integrations:
    • Crypto: Binance, Bybit, Hyperliquid
    • Stocks: Alpaca, Interactive Brokers
  • One-click deploy from paper trading to live
  • Kill switch — instant stop-all
  • Risk limits: max position size, daily loss limit, drawdown circuit breaker
  • Gradual scaling: start with small allocation, increase as confidence grows

6. Dashboard & Analytics

  • Portfolio overview: total value, P&L, positions
  • Strategy performance:
    • Sharpe ratio, Sortino ratio
    • Max drawdown, recovery time
    • Win rate, profit factor
    • Trade-by-trade log with entry/exit reasoning
  • Strategy comparison: side-by-side A/B view
    • Same market conditions, different strategies
    • Statistical significance testing
  • Charts: equity curve, drawdown chart, returns distribution

7. Market Data

Phase 1:

  • Historical OHLCV data (free sources: Yahoo Finance, CoinGecko)
  • Daily + hourly candles
  • Major crypto pairs + US stocks

Phase 2:

  • Real-time streaming data
  • Order book depth
  • Alternative data: news sentiment, social mentions, on-chain metrics
  • Premium data providers (Polygon.io, Kaiko)

Technical Architecture

┌─────────────────────────────────────────────┐
│                   Frontend                   │
│        React + TanStack Router + Tailwind    │
│  Dashboard │ Strategy Builder │ Backtest UI  │
└──────────────────┬──────────────────────────┘
                   │ tRPC + TanStack Query
┌──────────────────┴──────────────────────────┐
│            API Layer (tRPC)                   │
│                                              │
│  Strategies │ Backtests │ Portfolios │ Auth  │
│  Market Data │ Trade Log │ User Settings     │
└──────────────────┬──────────────────────────┘
                   │ Drizzle ORM
┌──────────────────┴──────────────────────────┐
│     Cloudflare D1 + Workers                  │
│                                              │
│  Backtest Engine │ Paper Sim │ Live Trading  │
│  LLM Gateway │ Data Fetcher │ Risk Manager  │
└─────────────────────────────────────────────┘

Stack:

  • Frontend: React, TanStack Router, TanStack Query, Tailwind, Recharts/Lightweight-Charts
  • API: tRPC
  • Database: Drizzle ORM + Cloudflare D1
  • Auth: Better-Auth + Polar (payments)
  • Strategy execution: Cloudflare Workers / Durable Objects
  • LLM integration: Multi-provider (Anthropic, OpenAI, Google via API keys)
  • Deployment: Cloudflare (full stack)

Data Model (Core)

Strategy
├── id, name, description
├── type: "algorithmic" | "llm" | "hybrid"
├── config: JSON (indicators, thresholds, prompts, model selection)
├── visualFlow: JSON (node editor state)
├── code: string (generated/custom Python)
├── createdAt, updatedAt
└── userId

Backtest
├── id, strategyId
├── market, timeframe
├── dateRange: { start, end }
├── config: { startingCapital, commission, slippage }
├── status: "pending" | "running" | "completed" | "failed"
├── results: { sharpe, maxDrawdown, winRate, totalReturn, trades }
└── equityCurve: [{ date, value }]

Trade (simulated or live)
├── id, strategyId, backtestId?
├── type: "buy" | "sell"
├── asset, quantity, price
├── timestamp
├── reasoning: string (especially for LLM strategies)
├── slippage, commission
└── pnl

Portfolio
├── id, userId
├── mode: "paper" | "live"
├── cash, totalValue
├── positions: [{ asset, quantity, avgPrice, currentPrice, pnl }]
└── strategyId

MarketData
├── asset, timeframe
├── candles: [{ open, high, low, close, volume, timestamp }]
└── source, lastUpdated

Monetization

Tier Price Features
Free $0 2 strategies, 10 backtests/mo, daily data, delayed quotes
Pro $29/mo Unlimited strategies & backtests, hourly data, paper trading
Trader $79/mo Everything + live trading, real-time data, priority execution

Revenue drivers:

  • Subscription tiers
  • Marketplace commission (if/when strategy sharing is added)
  • Data upsells (premium feeds)

Roadmap

Phase 1 — Foundation (MVP)

  • User auth + onboarding
  • Strategy CRUD (algorithmic type only)
  • Built-in indicator library (RSI, MACD, SMA, EMA, Bollinger)
  • Historical data ingestion (crypto + US stocks, daily candles)
  • Backtesting engine with basic metrics
  • Results dashboard with equity curve + trade log
  • Strategy comparison view

Phase 2 — AI & Visual Builder

  • LLM strategy type (multi-model support)
  • Hybrid strategy type
  • Visual drag-and-drop strategy builder
  • Paper trading simulator (real-time)
  • Hourly candle data
  • Advanced metrics (Sharpe, Sortino, Monte Carlo)

Phase 3 — Go Live

  • Broker integrations (Alpaca, Binance)
  • Live trading execution
  • Risk management system (kill switch, limits)
  • Real-time streaming data
  • Mobile dashboard

Phase 4 — Community

  • Strategy marketplace (share/sell strategies)
  • Leaderboards
  • Social features (follow traders, copy trading)
  • Alternative data sources

Competitive Landscape

Platform Algo AI/LLM Visual Builder Backtesting Multi-Market Simulation First
AlphaForge ✅ ✅ ✅ ✅ ✅ ✅
QuantConnect ✅ ❌ ❌ ✅ ✅ ⚠️
TradingView ⚠️ ❌ ❌ ⚠️ ✅ ❌
Cod3x ❌ ✅ ❌ ❌ ❌ (crypto) ❌
Composer ✅ ❌ ✅ ✅ ❌ (stocks) ⚠️

AlphaForge's moat: The only platform combining visual building + AI strategies + rigorous backtesting across multiple markets.

Success Metrics

  • MVP: Working backtest engine with 3+ built-in strategies, displayable results
  • Month 1: 50 beta signups from Product Hunt / Hacker News launch
  • Month 3: 500 users, 5% conversion to Pro
  • Month 6: Live trading beta, 1000+ users

Document version: 0.1 — Feb 13, 2026 Author: Martin Tielemans + Molt