நாலாயிர திவ்ய பிரபந்தம் — Naalaayira Divya Prabandham
A Next.js 15 web app for all 4000 pasurams of the twelve Azhwars, plus Ramanuja Nootrandhadi (108), in Tamil script, with:
- English translation and a word-by-word glossary for about 3,450 verses
- Tamil life stories of every Azhwar (
src/content/azhwars.ts)
- A full reader at
/read for all 4000 pasurams. It has adjustable text size, an English on/off toggle, "continue reading", and a progress count of verses read. Progress is stored in the browser.
- Downloadable PDFs: one per prabandham (Tamil + English) and the complete book (Tamil)
- Infographics for the whole collection: verses per Azhwar and prabandham, Divya Desams, avatars, the twelve names, Thiruvaymozhi panns, sangu and chakram mentions
- A LangGraph agent on every pasuram page that uses a Groq-hosted model, with a key each visitor brings (BYOK), to write a Tamil பொழிப்புரை, பதவுரை and விளக்கவுரை, and to produce emotion and theme meters, story context, characters, places, a short lesson and a quiz
Run
npm install
npm run dev # http://localhost:3000
AI commentary: bring your own Groq key
On any pasuram page, open 🔑 Groq API key (BYOK), paste a key from https://console.groq.com/keys, and click சரிபார்த்துச் சேமி (verify and save). The app checks the key against Groq's model list and loads the models into a picker. The default model is openai/gpt-oss-120b.
- The key is stored only in that browser's
localStorage.
- It is sent to this app's
/api/agent route in the x-groq-api-key header, used for that one request, and never written to disk or logged.
- Optionally, set
GROQ_API_KEY (and GROQ_MODEL) in .env.local to give visitors who have no key of their own a server-side fallback.
Without a key, the agent skips the model step and shows only the word-level analysis.
The agent (src/lib/agent/graph.ts)
START → retrieve → analyze ─┬─(API key)→ interpret → review ─┬─(issues, < 2 tries)→ interpret
└─(no key)──────────────→ compose ←┘
↓
END
| Node | Job |
|---|
| retrieve | Loads the verse, the verses before and after it, the prabandham and the Azhwar |
| analyze | Deterministic: grammar profile from the glossary, proper names, places, avatars, symbols |
| interpret | Calls Groq: strict JSON-schema output on gpt-oss models, JSON mode on others. The response is validated with the Zod schema |
| review | Checks the output is in Tamil and every field is filled. If not, it sends the issues back to interpret |
| compose | Merges everything into the infographic payload |
/api/agent streams each step as NDJSON so the UI can show the pipeline as it runs. Results are cached in data/ai-cache/.
Data (data/)
scripts/build_data.py builds the dataset from:
git clone https://github.com/UHH-Tamilex/NalayiratTivviyapPirapantam /tmp/ndp
# download the two xlsx files from NDP_Pann_Prediction/Data
python3 scripts/build_data.py --tei /tmp/ndp --pann <with_pann.xlsx> --nopann <without_pann.xlsx> --out data
Numbering follows the standard 1–4000 scheme. The two Thirumadals are split into couplets (78 + 148). Ramanuja Nootrandhadi is numbered 4001–4108. Places, avatars and symbols are detected by keyword matching, so their counts are approximate.
Neither source repository states a license. Check with the authors before you publish this data.
PDFs (public/pdf/)
Most PDF libraries misplace Tamil vowel signs (ெ, ே, ை), so the PDFs are printed from the reader pages with headless Chrome instead. The print stylesheet adds a cover page, page numbers and A4 layout. To regenerate them after changing the data or the layout:
NEXT_DIST_DIR=.next-pdf npm run build
NEXT_DIST_DIR=.next-pdf npx next start -p 3123 &
node scripts/build_pdfs.mjs --base http://localhost:3123 # all 26 files
node scripts/build_pdfs.mjs --base http://localhost:3123 --only thiruppavai
The script looks for Chrome in the usual locations, or in CHROME_PATH if you set it. The separate NEXT_DIST_DIR lets this production server run alongside npm run dev. Every reader page also has a print button (அச்சிடு), so visitors can save any section as a PDF from their browser.
