Sev Nightingale

The system is a three-agent pipeline. The Extraction Agent pulls from a Universal Data Layer that serves 39 configurable data points across 8 categories, including technical indicators (21 preprocessors via pandas-ta), real-time prices (Binance WebSocket, sub-second freshness), macro factors like VIX, DXY, and inflation, on-chain signals, social sentiment, breaking news, derivatives data, and AI research synthesized by third-party ACP agents. The whole thing is config-driven with parallel execution, and a shared Redis cache drives per-user cost down as the platform scales.

The Decision Agent routes to any of the 7 frontier model providers (Claude, GPT, Grok, Gemini, DeepSeek, Kimi, or Qwen) across three reasoning tiers (Economy, Standard, and Premium), applying the chosen model's reasoning to the user's natural-language strategy. The Trading Agent then executes on Hyperliquid with 228 perp markets, non-custodial via API wallets and up to 50x leverage, or simulates with a professional paper engine.

Users can opt into two competitive modes. The Dojo runs chess-style 1v1 Elo matches via copy-trade mirroring, while the Virtuals DGClaw Arena runs on-chain competition through the ACP protocol. The platform handles metered billing with Stripe and crypto payments via NOWPayments, Google OAuth authentication, and 5-second account monitoring, and every line of it was written by AI.

The stack is PostgreSQL with PostGIS for spatial queries, Apache Jena Fuseki as the RDF graph database for ontology-level reasoning, and geohash tile compression (6.46M tiles) that makes global occurrence lookups fast across 847 One Earth ecoregions. A Python Flask microservice handles OWL/RDF ontology work, Google Sheets ingestion, and the PostgreSQL to Fuseki knowledge-graph sync.

On top of that foundation I built L.E.A.F. (Location-based Ecological Aptness Forecast), a scoring mechanism that ranks candidate species for any given location and reforestation goal, runs fresh AI research on the top matches, and produces a restoration guide covering species selection, planting methodology, timing, and biodiversity strategy. I'm the technical lead at Silvi.

Sebastian runs on seven daily heartbeats, each tied to a room in a manor file architecture I designed: morning orientation, communications, observation, execution, diagnostics, accounting, and library curation. The system is built on Bun and TypeScript for the Telegram bot, Python for the heartbeat scripts, and the Claude Agent SDK for the agent runtime itself. MCP integrations cover Google Workspace (Gmail, Drive, Calendar, Sheets), Notion, Playwright, browser-use, Context7, and a custom X client for managing the @ggbots_ai and @sebastiansidoh accounts.

Concretely, Sebastian maintains Westlook Press for the writer Kristin Ellowe. She emails him with change requests, he triages them into Notion tasks, works them, and ships changes to Vercel autonomously. He ships NatureDataLab bug fixes from Telegram messages I send him. He manages my calendar and Notion task board, runs the @ggbots_ai X account, drafts articles for the blog, and sends emails to users. He runs a daily market conditions analysis that briefs me and any ggbot in the system on the current state of markets and macro. He reads Silvi team meeting notes, recaps priorities to the group Telegram, and updates our Notion tasks. He maintains a second brain and a personal CRM with files on people, and helps with day-to-day prioritization and long-term strategy. Occasionally he builds his own ggbots and runs trading strategy experiments.