Each entry is a typed JSON record in /content/, ready to be embedded into the Phase 2 retrieval index. The trust ranking surfaces curated entries first.
YouTube transcripts
Lectures and khutbahs from NAK, Mufti Menk, Assim al-Hakeem, Yasir Qadhi, Omar Suleiman. Abundant, free, transcription errors handled by manual review.
Books (PDF)
Mufti Taqi ʿUsmānī’s foundational works, AAOIFI Shariah Standards, classical fiqh manuals. Copyright handled under personal-use scope.
Web articles & fatwas
joebradford.net, amjaonline.org, islamicfinanceguru.com, and a small set of scholarly blogs. Indexed with permalink and capture date.
Hand-curated quotes
The high-trust core. Quotes I have personally verified against the source — slow to grow, never auto-generated.
The four buckets, ranked
| Bucket | Trust | Volume | Effort to add |
|---|---|---|---|
| Hand-curated | Highest | Smallest | High |
| Books (PDF) | High | Medium | Medium |
| Web articles | High | Medium | Low |
| YouTube transcripts | Medium | Largest | Lowest (auto) |
Phase 2 retrieval will weight these accordingly: a hand-curated quote outranks a YouTube transcript every time, even if both address the same question.
The pipeline
- Source identified — book, lecture URL, article URL.
- Captured — PDF saved locally; transcript scraped or fetched; article archived (with permalink + capture date).
- Indexed — chunked, embedded, stored. The index is private and not republished.
- Vetted — for the hand-curated bucket, I read the quote in full context and tag it with a confidence note.
- Retrieved — Phase 2 surfaces matches against questions with citation back to the original.