Social network intelligence: X trend analysis

Three layers built on the same local archive:

Scraper — pulls liked tweets off X, writes JSON / CSV / Excel / Markdown / HTML, downloads media. Resumes from a checkpoint after a network blip or rate-limit wait.
Search — hybrid recall (BM25 + dense embeddings, fused with RRF and re-ranked) over the archive. Reachable as a CLI or as an MCP server for Claude Code, Claude Desktop, and other MCP clients.
Social Network Intelligence — synthesis reports (brief / synthesis / trend) with a mermaid mindmap of authors, sources, themes, and hashtags. DSPy-typed prompt with an entity-relevance filter; in-memory knowledge graph you can dump as mermaid or JSON.

Next Projects

AI Skills for strategic foresight, intelligence analysis and investigation theory

May 27

AI Skills for decision intelligence and visualization.

Nvim Simple to reduce the AI vs. human context gap when you take over the work

Feb 11

Nvim Simple is my version of a different Nvim usage. It's optimized for taking over when the AI agents fail.

Terminal for Automation with Quant Models

Aug 29

Quant terminal for quantitative financial models. No AI involved. Math and Python with thorough testing for hypothesis building.

RAGflow MCP server for Claude Desktop and Agentic Coding

Jun 10

RAGflow is a knowledge base for your books, reports, documents, and tables. It uses custom (local) models and established (remote) LLMs to interact with AI agents. I wrote an MCP server for it, that works with Claude, opencode and other clients