Assalamu alaikum seemorg team,
Cross-posting from email since contact@usul.ai bounced (Cloudflare email routing returned 421 4.3.0 upstream error, Gmail retried for 22 hours then gave up). FYI in case other people are seeing the same issue.
Context
I'm Faiz Mohd, founder of HalalHQ (https://halalhq.io), an Australian halal lifestyle platform with a Hajj and Umrah feature, mosque directory, prayer times, and a halal-product database with barcode scanner.
We are scoping a strict RAG-only Islamic Q&A surface (citations always visible, scoped to source-text Q&A and explicitly out-of-scope for fatwa-level rulings). Waleed Kadous from the Ansari project (https://ansari.chat) pointed us to usul.ai for the retrieval layer over Shamela / classical scholar texts. After reviewing your indexing depth, the seemorg/usul-data MIT licence, and your OpenITI + Turath.io ingestion, we would much rather call your hosted API than re-host the corpus ourselves.
Three things we could not find documented anywhere public
- Hosted API access: is
api.usul.ai available for third-party developers? If so, what is the auth model (API key, OAuth)?
- Rate limits / hosted-API pricing: published tiers, usage-based, or case-by-case? We would be at under 10K queries per month at launch.
- AU-region latency: where is the vector DB / inference hosted? Australian users are a hard requirement.
One corpus question
The usul-data README confirms Shamela is incorporated via OpenITI + Turath.io. Does the hosted API surface Mawsuah al-Fiqhiyyah al-Kuwaitiyyah specifically as a subset of that, or is it a separate ingestion outside scope? Mawsuah is the bulk of our fiqh-corpus need.
Why we are committing to this in public
We have a hard launch gate before any Islamic Q&A surface ships, internal alpha included: 100% pass on BATIK (Ansari, in hand), zero hallucinated text, zero hallucinated references. Mirroring Waleed's standard. Happy to publish our scores when we ship.
Even directional answers help us scope. Fall-back is self-hosted pgvector indexing of Shamela ourselves, but we would much rather not duplicate work you have already done.
JazakAllahu khayran for the work, particularly for keeping the dataset MIT-licensed.
Wassalām,
Faiz Mohd
Founder, HalalHQ.io
Assalamu alaikum seemorg team,
Cross-posting from email since
contact@usul.aibounced (Cloudflare email routing returned421 4.3.0upstream error, Gmail retried for 22 hours then gave up). FYI in case other people are seeing the same issue.Context
I'm Faiz Mohd, founder of HalalHQ (https://halalhq.io), an Australian halal lifestyle platform with a Hajj and Umrah feature, mosque directory, prayer times, and a halal-product database with barcode scanner.
We are scoping a strict RAG-only Islamic Q&A surface (citations always visible, scoped to source-text Q&A and explicitly out-of-scope for fatwa-level rulings). Waleed Kadous from the Ansari project (https://ansari.chat) pointed us to usul.ai for the retrieval layer over Shamela / classical scholar texts. After reviewing your indexing depth, the
seemorg/usul-dataMIT licence, and your OpenITI + Turath.io ingestion, we would much rather call your hosted API than re-host the corpus ourselves.Three things we could not find documented anywhere public
api.usul.aiavailable for third-party developers? If so, what is the auth model (API key, OAuth)?One corpus question
The
usul-dataREADME confirms Shamela is incorporated via OpenITI + Turath.io. Does the hosted API surface Mawsuah al-Fiqhiyyah al-Kuwaitiyyah specifically as a subset of that, or is it a separate ingestion outside scope? Mawsuah is the bulk of our fiqh-corpus need.Why we are committing to this in public
We have a hard launch gate before any Islamic Q&A surface ships, internal alpha included: 100% pass on BATIK (Ansari, in hand), zero hallucinated text, zero hallucinated references. Mirroring Waleed's standard. Happy to publish our scores when we ship.
Even directional answers help us scope. Fall-back is self-hosted pgvector indexing of Shamela ourselves, but we would much rather not duplicate work you have already done.
JazakAllahu khayran for the work, particularly for keeping the dataset MIT-licensed.
Wassalām,
Faiz Mohd
Founder, HalalHQ.io