A Shia hadith research platform. 32,519 narrations from 18 books, searchable in Arabic and English, cross-linked to similar narrations and to the Quranic verses that illuminate them.

Live at hadith.academyofislam.com.

Spring Boot 3.3.2 on Java 17, with Elasticsearch as the only datastore. Everything expensive — semantic embeddings, similar-narration judgments, hadith-to-Quran connections — is computed offline by Python scripts and Claude sub-agents, then stored in the index. No LLM is called while serving a request.

You need Java 17, Maven, and Elasticsearch 9.x on localhost:9200 with the corpus already indexed. Without a populated index the app starts but every search returns nothing; see docs/data-pipeline.md to build one.

mvn -B package -DskipTests          # build
./scripts/ops/restart.sh            # run with the dev profile, detached

restart.sh kills whatever holds port 8002 and relaunches detached, so the JVM survives your shell. To run in the foreground instead:

mvn spring-boot:run -Dspring-boot.run.profiles=dev

Spring DevTools does not reliably pick up CSS changes — restart to verify styling.

# what CI runs
mvn test -Dtest='!*IntegrationTest, !HodaAlQuranQualityCheck' -DfailIfNoSpecifiedTests=false

mvn test                            # everything, needs a live Elasticsearch
node --test src/test/js/*.test.js   # frontend unit tests

*IntegrationTest needs a running Elasticsearch; HodaAlQuranQualityCheck scrapes hodaalquran.com to verify extraction quality. Both are excluded in CI.

src/main/java/com/rewayaat/
  controllers/       Server-rendered pages — the crawlable surface
    rest/            JSON API — what the Vue app talks to
  service/           Business logic
  core/              Query building, result shaping, text processing
    data/            Persisted models (HadithObject, QuranVerse, ...)
  tafsir/            Tafsir parsing; extractors/ holds 13 source-specific parsers
  loader/            One-time corpus loaders, per book
  tools/             Runnable offline backfill and audit tools
  config/            Spring configuration
src/main/resources/
  templates/         Thymeleaf pages
  static/            CSS, JS, taxonomy.json, quran.json, book_blurbs.json
src/test/            JUnit tests, plus node:test frontend tests in js/

scripts/             Offline pipelines, grouped by purpose — see scripts/README.md
batches/             The source corpus as JSONL, imported into Elasticsearch
docs/                Architecture, search, pipelines — see docs/README.md
k8s/                 Kubernetes manifests
.github/workflows/   CI/CD
tmp/                 Symlink to /mnt/share/rewayaat-backup/tmp — large artifacts, never committed

Defaults live in src/main/resources/application.yaml, overridden per profile by application-dev.properties and application-prod.properties. Everything is environment-driven:

Both the dev and prod profiles point Quranic insights at rewayaat_quranic_light_filtered. The bare application.yaml default is the unfiltered rewayaat_quranic_light, so always run with a profile.

Hadith edit access is a static email allowlist in src/main/resources/admins.txt.

Pushes to master that touch src/**, pom.xml, the Dockerfile or the workflow build a Docker image, push it to Docker Hub, and update the image tag in k8s/kustomization.yaml. Argo CD picks that commit up and rolls it out to DigitalOcean Kubernetes. Pull requests run the build and tests but never deploy.

• The canonical host is hadith.academyofislam.com. The app also answers on rewayaat.info, which is a mirror. BASE_URL in HomeController is deliberate — do not "fix" it.
• manuscript.css has duplicate blocks by design — base styles, then an ornament pass that overrides them. Do not consolidate them.
• tmp/ is a symlink to another disk. The main disk runs near full; large artifacts must go through tmp/.
• Never git checkout HEAD a file with uncommitted changes. Several pipelines write into the working tree.

See LICENSE.