Hey PH 👋 I built Alpstein because most crypto "analysis" I came across fell into two unsatisfying buckets — noise-heavy news aggregators, or technical chart tools without context. Neither one does what a human analyst actually does: read the news, cross-reference current price action, weigh the two, and form a position with a defined entry, target, and risk. I wanted to see whether an LLM could do a reasonable version of that loop end-to-end, and whether it could be packaged as something anyone could use. Under the hood: a Node.js scraper pulls articles from four crypto news sources, pushes them through a Redis queue to a Go LLM service. For each article, the model calls OpenAI with two tool calls available — one for live Binance market data, one for RAG retrieval against recent articles stored as embeddings in Qdrant. With both grounded, the model produces a structured trade opinion (long/short, entry, stop-loss, take-profit, rationale). A separate WebSocket service tracks live prices and computes real-time P&L for each open position. Built solo over the last several months. Hosted on Hetzner + Vercel. The whole pipeline is instrumented with Prometheus, Grafana, OpenTelemetry, and Jaeger which made prompt tuning much easier than I expected. Genuinely curious what people think, especially about the agent design, RAG approach, and whether the trade-opinion framing lands. Feedback welcome, including the harsh kind.

Alpstein

AI-powered crypto news analysis with live P&L tracking.

AI-powered crypto news analysis with live P&L tracking.