A staged method, not a single answer
Taaweel breaks a research question into stages that can each be inspected, repeated and corrected. The pipeline runs end to end with live progress, and you can stop, resume or rerun a stage with different settings.
What it does
Quranic inquiry rarely starts as a clean question. The first stage is a conversation: the refiner restates the question, narrows its scope and asks what is missing. Only then does the engine begin.
From there the work follows the structure of a research paper. It plans, maps the corpus, collects evidence, builds a model of the answer, attacks it, revises it and composes the report. Each stage stores its output, its jobs, its token use and its cost, so nothing is a black box.
Reproducibility, stated plainly
Language models are probabilistic. Taaweel records the configuration of each run so results can be compared, but two runs on the same question may differ. Treat the pipeline as a way to structure and check work, not as a deterministic oracle.
- Sessions, runs, events and artifacts are persisted and can be revisited.
- Reports export to Markdown, HTML and DOCX.
- Corpus databases are opened read-only, so research cannot alter the sources.
Step by step
Plan
Question routing and a research design with an explicit warrant and expected hard cases.
Cartograph
Whole-corpus root and string concordance with clustering.
Collect
Tiered evidence from verse, lexicon, tafsir and hadith.
Audit
Objections ruled on; principles checked; revisions applied.
Compose
Citation-first report with appendices.
What you get
Live progress
Stage events stream into the interface as they happen, so long runs stay observable.
Stop and resume
Cancel a run safely and pick it up later; completed stages are not repeated.
Per-stage models
Choose the language model for each stage independently, including provider routing across OpenAI, DeepSeek, Z.AI and OpenRouter.
Spend control
A monthly cap on AI spend with per-job cost tracking.
Common questions
How long does a run take?
It depends on the question, the models chosen and how many passages are collected. Runs are designed to be long-lived background jobs with progress you can watch.
Can I rerun just one stage?
Yes. Stages are stored separately, so a stage can be rerun with different models or settings without redoing the whole run.
Does the pipeline need an internet connection?
The corpora are local and read-only. Model calls go to the provider you configure.
See Taaweel run on your question
Tell us the question you are researching. We will walk through a run and what the evidence ledger and audit look like.