Quran Bundle Search

Bundles

The Quran is partitioned into overlapping windows of N consecutive ayahs, called bundles. Every possible N-ayah block is indexed — for example, with N=4 and a surah of 8 ayahs, the bundles are ayahs 1–4, 2–5, 3–6, 4–7, and 5–8. The bundle size N (1–5) is adjustable in the header; changing it rebuilds the index. A larger N captures broader thematic context; a smaller N is more precise.

Must Exist & Must NOT Exist

Boolean filters over the full bundle index. As you type Arabic, the autocomplete resolves your input to either a root (trilateral Quranic root — matches every derived word form across the Quran, using the Leeds Morphology Corpus) or an exact word form (up to 20 tags total). All Must Exist terms are ANDed; Must NOT Exist terms are subtracted. Matching is morphologically normalised (diacritics, hamza, alef variants collapsed) but otherwise exact — no fuzzy or phonetic matching.

Semantic Sort

Provide a concept phrase in English or Arabic (up to 200 characters). Each bundle's Arabic text is embedded into a 384-dimensional vector using paraphrase-multilingual-MiniLM-L12-v2 — a cross-lingual sentence-transformer that maps Arabic and English into the same semantic space. Results are re-ranked by cosine similarity between your query embedding and each bundle embedding. Embeddings are pre-computed and cached; ranking is sub-second. Semantic sort re-ranks the results of the boolean filters — it does not replace them.

Result Analysis

Shown automatically when results are 200 or fewer bundles. For each term (root or word) across the result set, the analysis computes a DF ratio:

DF ratio = (freq in results ÷ result count) ÷ (freq in Quran ÷ total bundles)

A ratio of 5× means a term is 5 times more concentrated in your results than in the Quran at large — a signal of thematic relevance. Universal terms appear in every result bundle. Distinctive lists all terms sorted by DF ratio descending. Stopwords can be toggled off.

Tips

Use roots for thematic sweeps (e.g. ر‌ح‌م covers رحمة, رحيم, رحمن, …); use word forms to pin exact vocabulary.
Combine Must Exist + Semantic Sort: boolean filters constrain the candidate set, semantic sort surfaces the most conceptually aligned bundles within it.
The Bundle Step control (appears after search) lets you navigate overlapping bundles, skipping by step size to reduce redundancy.

کاشف