Projects
Public experiments in evidence-grounded agent workflows, reproducible analysis and reliable knowledge systems.
Pallas Ads on GitHub
An independent, free and open-source plugin for doing advertising analysis inside Codex and Claude Code.
Analysis, competitor research and dashboard outputs. The advertising performance examples use synthetic data. Click to enlarge. Pallas saves confirmed product profiles and guides the agent through a repeatable analysis workflow. Business context carries forward, so users can ask follow-up questions without repeating the same background and analysis requirements.
Supported account connections or CSV/XLSX exports supply the data for traceable HTML diagnostic reports, competitor analyses and local dashboards. Historical data stays in the workspace for reuse, and the host agent can schedule dashboard updates.
Agentic Knowledge Vault on GitHub
A portable Markdown vault that turns long transcripts and web captures into knowledge people and agents can reuse.
The knowledge workflow: human source selection, a curated Markdown vault and agent-assisted outputs. Click to enlarge. Raw evidence is kept separate from source notes and concept notes. Exact source paths, YAML metadata and topic indexes help people and agents find relevant material and trace each note back to its origin.
People decide what belongs in the vault and judge whether a note is faithful. Agents handle formatting, links and indexes under explicit project rules. The public demo follows two sources from capture to processed notes.
Jieyi Skills on GitHub
Reusable skills for the two ends of a knowledge workflow: capturing sources and studying what you keep.
Python scripts capture public X posts and YouTube transcripts as immutable raw sources with consistent names and metadata. The agent then handles the synthesis and maintains the processed notes.
A separate workflow turns vetted notes into a self-contained HTML study page with search, filters and flashcards. The output works offline and can be read without Obsidian.
