Searchable Learning Ledger and Opportunity Finder

Constraints

Operating boundaries

• Run local-first with no LLM calls, no external APIs, and no secrets printed.
• Treat rankings as decision support, not final authority.
• Cross-check candidates against completed local artifacts before reporting them as active work.
• Keep completed work visible for auditability instead of hiding it.

Build

What I built

• A search utility for querying indexed source learnings by keyword and artifact metadata.
• An opportunity finder that scores repeated themes across ledgers, demos, skills, and scripts.
• A completion cross-check that separates active opportunities from already implemented capabilities.
• Regression tests to prevent completed opportunities from being ranked as active next work.

Architecture

Workflow

• Data sources: learning-ledger JSON, portfolio-demo Markdown, local skills, PA scripts, and generated index artifacts.
• Scoring: category templates plus keyword evidence, source count, and readiness signals.
• Completion detection: local artifact markers such as scripts, tests, indexes, and reference docs.
• Outputs: timestamped JSON and Markdown reports with Active Opportunities and Completed / Already Implemented sections.

Security / Operations

Key decisions

• The finder is deterministic because this is an ops hygiene tool, not a creative writing task.
• Completed opportunities are still reported, but demoted and labeled with evidence so the exclusion is auditable.
• The tool favors class-level opportunities over narrow per-video actions to avoid brittle one-off automation sprawl.

Impact

What changed

• Turned passive learning artifacts into an implementation queue.
• Caught and corrected a failure mode where completed work was being presented as new work.
• Improved trust by making recommendation state inspectable: active, completed, evidence, and next step.

Evidence

Source trail