tomjnet

Premarket AI — Data skill for Claude Code

Data community

AI pre-market news platform that separates real market news from fake.

How to install Premarket AI

This entry records only its repository, not the path inside it, so there is no exact command to give. Open tomjnet/premarket-ai and copy the folder into ~/.claude/skills/, or the file into ~/.claude/agents/.

What Premarket AI does

AI pre-market news platform that separates real market news from fake. It migrates a legacy C++11/PostgreSQL/PDF pipeline to C++20, RAG, LangGraph agents, MCP, Skills and local LLMs (with OpenAI/Claude/Gemini switch), all on Podman(Docker).

Alternatives in Data

Private GPT — Complete API layer for private AI applications on local models: RAG, skills, tools, MCP, text-to-sql, and more 57.5k ★
OpenViking — Self-evolving Context Database for AI Agents 33.3k ★
Local News Site — Publish a local news article as a standalone Elementor page, with Google News-grade structured data and image 648 ★

README

premarket-ai

AI pre-market news platform that separates real market news from fake. It migrates a legacy **C++11 / PostgreSQL / PDF** pipeline, step by step, to **modern C++20, RAG, LangGraph agents, MCP, Skills and local LLMs** (with an OpenAI / Claude / Gemini switch), all running on **Podman**.

⚠️ **Simulation, not investment advice.** All vendor news in this project is **synthetic**: it's generated, labeled `[SYNTHETIC]`, and fake sources use reserved `.example` / `.test` domains. The system is decision support only. It never recommends buying or selling and has no trading tools.

The story

Before the market opens, traders read a news report. For years it came from a **batch pipeline**: an external vendor sends ~100 news items per day, a C++ program loads them into PostgreSQL, and another C++ program prints a **PDF** onto an NFS share.

The vendor is paid for 100 items per day, so it **pads the feed with duplicates, stale stories, fake news and even fake companies**. After a conference, the CEO asked for a modern website and for AI that can tell traders **which news is real**.

This repo rebuilds the system **one increment at a time** (the strangler-fig pattern). Every increment ships something traders can use and replaces one piece of the legacy system, until the PDF is retired.

Delivery increments

Each increment is developed on its own branch, merged to `main` through a PR once its "done when" check passes, and tagged (`inc-0`, `inc-1`, …) so every stage can be checked out exactly as it was.

#	branch_name	Traders get
0	`inc-0-legacy-baseline`	The PDF report, as today (the "before")
1	`inc-1-web-instead-of-pdf`	A news feed website instead of the PDF, no AI yet
2	`inc-2-rule-based-checks`	FAKE COMPANY / DUPLICATE / STALE / SPOOFED SOURCE badges from rules, no LLM
3	`inc-3-first-ai`	AI summaries, sentiment and "Ask the News" chat with cited sources
4	`inc-4-ai-verification`