WeBuddhist Creator Assistant — Bodhisattva Challenge

A chat-assistant website that helps Buddhist content creators make short videos for The Bodhisattva Challenge — a 365-day journey through Śāntideva's Bodhicharyavatara.

The creator picks a day; the app pulls that day's verses, plan, and classical commentary from a local clone of the source repo, gives a simple plain-language summary of the verse (in English or Hindi), suggests the video ideas that day's material can support, and generates either a ready-to-read script or a shot-by-shot video structure. Results can be regenerated, refined by chatting, and (for scripts) turned into narrated audio.

Stack

• Frontend: React (Vite) — guided chat UI.
• Backend: Django + Django REST Framework.
• LLM + TTS: Google Gemini (one API key for both text and audio).
• Source content: a local clone of the bodhisattvacharyavatara-rails repo.

Architecture

frontend (React/Vite)  ──HTTP──>  backend (Django REST)  ──reads──>  rails repo (markdown)
                                          │
                                          └──> Gemini API (summary · script · structure · audio)

The backend reads the day-plan and per-verse commentary files directly from disk (no database for app data — generated content is ephemeral and flow state lives in the browser). It runs one Gemini call to decide which ideas a day supports, and fills editable prompt templates ("skills") to generate everything else.

The creator flow

1. Pick a day (1–365).
2. See the day — a Day / Chapter / Verses / Date strip plus today's verse text.
3. Verse summary — choose English or हिन्दी Hindi for a few short, plain-language bullet points explaining the verse; optionally get the other language too.
4. Pick a video idea (see below).
5. Choose the output — 📝 Video script (spoken script) or 🎬 Video structure (a storyboard: timed beats with on-screen visuals + voiceover).
6. Pick a duration — 30 / 45 / 60 / 90 seconds.
7. Get the result, then ↻ Regenerate for a fresh take, or chat to refine it ("make the hook punchier", "shorten the opening"). Scripts can also be turned into narrated audio.

Video ideas

Idea	What it does	When offered
Concept	Explains the core idea of the verse	always
Challenge / Practice	Frames today's practice as a doable dare	always
Creative	A secular, universal video about the lesson behind the verse — no scripture or Buddhist references, crisp one-liners, for everyone	always
Testimony	Shapes the creator's own notes into a first-person reflection	always
Story	Tells a story/parable from the source	when the material contains one
Extra info / fun fact	A surprising detail from the texts	when the material contains one

Local setup

Quick start

make install        # backend venv + deps, frontend npm install
make dev            # runs backend (:8001) and frontend (:5173) together; Ctrl+C stops both

Backend (manual)

cd backend
python3 -m venv venv
venv/bin/pip install -r requirements.txt
cp .env.example .env        # then edit .env

Set in backend/.env:

• RAILS_REPO_PATH — absolute path to your local clone of bodhisattvacharyavatara-rails.
• GEMINI_API_KEY — your Google Gemini API key (summary, script, structure, audio).

venv/bin/python manage.py migrate          # built-in tables only
venv/bin/python manage.py check_content    # verify it can read the rails repo
venv/bin/python manage.py runserver 0.0.0.0:8001

Frontend (manual)

cd frontend
npm install
cp .env.example .env        # VITE_API_BASE_URL defaults to http://localhost:8001
npm run dev                 # http://localhost:5173

Open http://localhost:5173 and start with a day number (e.g. 5).

Without a GEMINI_API_KEY, day loading still works and idea availability falls back to a keyword heuristic — but the verse summary, script, structure, and audio all require Gemini and return a clear "not configured" message.

API

Method	Path	Purpose
GET	/api/health/	status + whether Gemini is configured
GET	/api/days/<n>/	verses, chapter, date, verse text, available ideas
POST	/api/verse-summary/	{day, language} → {points} — language is english or hindi
POST	/api/script/	{day, ideaKey, durationSeconds, creatorNotes?, feedback?, previous?} → {script}
POST	/api/structure/	{day, ideaKey, durationSeconds, creatorNotes?, feedback?, previous?} → {structure}
POST	/api/audio/	{script, voice?} → {audioUrl}

feedback + previous power the chat-to-refine step (revise the prior result); omit them for a fresh generation.

The "skills" (prompt templates)

Every generation is driven by an editable markdown template in backend/assistant/prompts/ — tune tone, structure, and rules without touching code (changes take effect on the next request, no restart needed):

File	Role
_shared.md	Shared, source-faithful context + voice rules for scripts
concept.md, practice.md, testimony.md, story.md, extra_info.md	Per-idea script angles
creative.md	Self-contained Creative script (secular, no source references)
verse_summary.md	Simple plain-language verse summary (English/Hindi)
structure.md	Source-faithful video storyboard
structure_creative.md	Secular, universal storyboard for the Creative idea
assistant.md	Greeting / persona

Security & limits

• Rate limiting — the Gemini-backed endpoints are throttled (default 20/min for generation, 60/min elsewhere; tune via THROTTLE_GENERATE / THROTTLE_ANON in .env) to protect the API budget.
• Input caps — creatorNotes, script, and feedback are length-capped.
• Production guards — with ENV=production, the app refuses to start if DEBUG is on or DJANGO_SECRET_KEY is left at the insecure default.
• Caching — day content and idea analysis are cached in memory; identical audio requests reuse the existing file. Verse summaries are cached in production and regenerated live in local dev.

Deployment

See DEPLOY.md for EC2 (gunicorn + nginx).