Choose a color theme
GitHubData engineering · Airflow · Python
This project treats personal AI and search history as source material for a writing workflow. The aim is not to publish a raw activity dump; it is to move from messy exported records to useful, reviewable technical writing.
The problem
AI conversations and search history contain valuable questions, dead ends, and discoveries, but they are noisy and difficult to reuse. A manual process loses chronology and makes it hard to tell whether an article reflects evidence or memory.
What I owned
- The Airflow-oriented workflow and task boundaries.
- Python transformations for exported history and article-ready records.
- Validation steps that preserve source context and surface incomplete data.
- A path from raw history through review to a publishable Markdown article.
Design decisions
The pipeline separates extraction, normalization, analysis, drafting, and review. That makes each stage inspectable and gives an editor a place to intervene before publication. The workflow is intentionally compatible with the same Markdown-based VitePress archive this site publishes.
Evidence and outcome
The project turns an otherwise transient development trail into durable documentation. It also demonstrates a practical data-engineering principle: automation becomes more trustworthy when intermediate representations and validation are visible.