Sanad

The problem

For Muslims, the hadith, the recorded sayings of the Prophet Muhammad ﷺ, are the second most important body of text after the Quran. Students and researchers ask the same questions of each one. Who passed it on, and how reliable was each person in that chain? Where do two chains for the same hadith meet or split? Which narrators sit at the centre of the whole network?

Sites such as sunnah.com are good readers, but they offer almost nothing for these questions. Sanad is for Muslim students, general readers and Islamic-studies researchers who want to read a hadith with its chain and its grading in view.

One rule shapes the whole product. The app never issues its own religious verdicts. Arabic text is shown exactly as sourced, and every reliability grade is attributed to the classical scholar who gave it.

What I built

I compute the analytics ahead of time

Counting words, comparing texts and ranking narrators is heavy work across tens of thousands of hadith. The collection itself barely changes. So I run all of it in advance, in background jobs, and store the results. When someone opens a page, the server only reads rows that already exist. Everything else in the system builds on that decision.

Chains are pulled out of the Arabic text

The source dataset has the hadith text but no structured list of narrators. I wrote a parser that finds the chain in each Arabic text by its transmission phrases, such as "narrated to us" and "from", and pulls out the names. From those names it builds a network of teacher and student links. The app labels this as a structural estimate and shows where it came from. Clean names took many rounds of fixes. Fragments of text that aren't names are dropped. Unnamed narrators (mubham) are grouped under one placeholder, and the app never invents an identity for them.

Each graph size gets its own renderer

One hadith's chain is a handful of people. It renders as an interactive graph, and clicking a narrator opens their profile. The whole narrator network runs to hundreds of people and more, which is too many for that tool. So it renders on a canvas with a force layout, with search, focus and a shortest-path finder between any two narrators. I wrote that split into the spec from the start.

Every grade names its source

Each grade shows the scholar who issued it, and a hadith can carry grades from several scholars side by side. Narrators without an assessment show "not yet assessed" in a neutral colour that can't be read as a verdict. I reworked the grade colours once, after an early palette gave some colours meanings they shouldn't have had.

I gave finished backend work a screen

Midway through, I found six analytics endpoints that worked but had no page. I spent the next sprint building screens for them. The grade-distribution explorer, the word-frequency tool, the corpus dashboard, the parallel-narrations panel and the narrator ranking all shipped in that wave.

Result

Sanad is feature-complete and staged for launch, but not public yet. Its data depends on an ingestion pipeline I run locally. Readers get the Arabic text with vowel marks, English and Indonesian translations side by side, grade badges, a focus mode and resume-reading. Signed-in users get bookmarks, collections and reading history. Any hadith can be exported as an image card to share. The original plan had AI topic search. I shipped a curated topic browser over full counts, with no paid AI service behind it. I built Sanad in five days in June 2026, and it grew well past its original spec.