ComHedge

Why I Built This

Commodity trading is still stuck in the dark ages. Traders are manually scanning EIA reports, CFTC positioning data, weather forecasts, and news feeds - then trying to synthesize all of it into a trade thesis. That’s insane. Machines should be doing the synthesis. Humans should be making the decisions.

ComHedge plugs into 12 real-time data feeds and runs them through multiple AI models to generate scored trading advisories. Not predictions - actionable trade ideas with confidence levels, bull/bear cases, risk factors, and verifiable sources.

What It Does

The platform generates 100+ trade ideas daily across crude oil, natural gas, corn, wheat, coffee, gold, copper, and more. Each advisory gets a confidence score from 0-100 based on consensus from Claude, GPT, Gemini, and Perplexity running in parallel. The system includes paper trading, P&L tracking, a leaderboard, and a full admin dashboard for monitoring data feed health.

16+ commodities. 12 data sources. 4 AI models. Real-time.

How It’s Built

• Framework: Next.js 15 with App Router and React 19
• Hosting: Cloudflare Workers via OpenNext adapter
• Database: Cloudflare D1 with Prisma ORM for advisory versioning and audit trails
• AI: Claude (primary generation), OpenAI, Gemini, Perplexity (second opinions)
• Data Feeds: EIA API, FRED, CFTC COT reports, NOAA weather, Baker Hughes rig count, Yahoo Finance, Alpha Vantage, NewsAPI, Grid Status
• Auth: Clerk for user management
• Charts: Recharts for P&L and performance visualization

The Architecture

This is the most technically ambitious thing I’ve built. The advisory generation pipeline works like this:

1. Data aggregation from 12+ authoritative sources
2. Hierarchical prompt system - category to subcategory to commodity-specific context
3. Intelligence briefing formatted with all relevant data
4. Claude generates the advisory with full analysis
5. Second opinions from 3 other LLMs in parallel
6. Consensus scoring and final publication
7. Database versioning for complete audit trail

The multi-LLM consensus system is the secret sauce. No single model’s bias dominates. When Claude, GPT, Gemini, and Perplexity all agree on a trade direction with high confidence - that signal is worth paying attention to.

Who It’s For

Commodity traders who want data-backed signals. Hedge funds managing commodity exposure. Energy companies hedging risk. Prop trading shops that need real-time market intelligence. Anyone trading futures who’s tired of gut-feel decisions.

FAQ

Is this live trading? Paper trading right now with a path to autonomous execution. The D2 Auto-Invest system is being tested on natural gas futures with capped bankroll and human-in-the-loop approval.

How accurate are the advisories? The leaderboard tracks every advisory’s outcome. Win rates, P&L, and performance attribution are all transparent. The multi-LLM consensus approach consistently outperforms any single model.

Who’s Dennis? A domain expert with 25+ years of commodity trading experience. His trading playbook and rules are being codified into the system. The AI learns from decades of human expertise.