🧠 DocuMind — AI Document Hub

A high-end, production-ready RAG SaaS for document management, semantic search, and contextual AI chat.

---

Overview

DocuMind is an advanced Retrieval-Augmented Generation (RAG) platform built for precision and scale. Upload PDF documents, extract and chunk text with LangChain, generate embeddings with Google Gemini, and store vectors in Supabase pgvector. A dark-themed split-view chat interface streams accurate answers with page citations.

The platform includes a complete SaaS monetization layer with Stripe webhooks, subscription-tier enforcement in Next.js middleware, and a full document management library.

---

Tech Stack

Layer	Technology
Framework	Next.js 15 (App Router, Turbopack)
UI	Tailwind CSS v4, Framer Motion, shadcn/ui, OKLCH colors
AI Embeddings	Google Gemini gemini-embedding-001 (768-dim) via @google/genai
AI Chat	Google Gemini gemini-2.5-flash-8b via Vercel AI SDK
Chat Hook	@ai-sdk/react useChat (AI SDK v6+)
Vector DB	Supabase pgvector with HNSW index (cosine similarity)
File Storage	Supabase Storage (auto-created public bucket)
PDF Parsing	LangChain WebPDFLoader + pdf-parse@1
Text Chunking	LangChain RecursiveCharacterTextSplitter (1000 chars / 200 overlap)
Payments	Stripe Checkout + Webhooks

---

Project Structure

src/
├── app/
│   ├── page.tsx                  # Upload hub + live document library + Recent Insights
│   ├── actions.ts                # Server actions: processPDF, fetchDocuments, deleteDocument
│   ├── layout.tsx                # Root layout + Sonner toaster
│   ├── globals.css               # Tailwind v4 + OKLCH design tokens (dark mode)
│   ├── hub/[id]/
│   │   ├── page.tsx              # Document chat page (server component)
│   │   └── ChatWorkspace.tsx     # Split-screen: PDF iframe + Gemini chat (client)
│   ├── documents/page.tsx        # Full document library grid (server)
│   ├── chat/page.tsx             # Document picker → chat workspace (server)
│   ├── api-reference/page.tsx    # REST API documentation (mock)
│   ├── settings/page.tsx         # Settings: Profile, API Keys, Storage, Security (mock)
│   ├── pricing/page.tsx          # Stripe subscription tiers
│   └── api/
│       ├── chat/route.ts         # Streaming RAG chat endpoint
│       └── stripe/               # Stripe webhook + checkout handlers
├── components/
│   ├── TopNav.tsx                # Shared navigation bar (active-route aware)
│   └── ui/                       # shadcn/ui primitives
├── lib/
│   └── supabase.ts               # Supabase anon client + service-role admin client
└── middleware.ts                  # Route guard (protects /api/* except Stripe webhook)

---

Pages

Route	Description
/	Upload zone + live document library + Recent Insights panel
/documents	Full document grid with status badges, word count, timestamps
/chat	Document picker list — click to open chat workspace
/hub/[id]	Split-screen: PDF viewer (left, 60%) + Gemini AI chat (right, 40%)
/api-reference	Mock REST API docs with expandable endpoints and copy-able curl examples
/settings	Mock settings: Profile, API Keys, Storage usage bar, Security toggles, Notifications
/pricing	Stripe-powered Free vs Pro subscription page

---

Database Schema

Run supabase_schema.sql in the Supabase SQL Editor before first use.

-- Required extension
CREATE EXTENSION IF NOT EXISTS vector;

-- Source documents
CREATE TABLE documents (
  id          UUID DEFAULT gen_random_uuid() PRIMARY KEY,
  file_name   TEXT NOT NULL,
  file_url    TEXT NOT NULL,
  content     TEXT NOT NULL,
  created_at  TIMESTAMPTZ DEFAULT timezone('utc', now()) NOT NULL
);

-- Vector chunks
CREATE TABLE embeddings (
  id           UUID DEFAULT gen_random_uuid() PRIMARY KEY,
  document_id  UUID REFERENCES documents(id) ON DELETE CASCADE NOT NULL,
  content      TEXT NOT NULL,
  metadata     JSONB DEFAULT '{}',    -- { loc: { pageNumber: N } }
  embedding    VECTOR(768) NOT NULL,
  created_at   TIMESTAMPTZ DEFAULT timezone('utc', now()) NOT NULL
);

-- HNSW index for fast cosine similarity search
CREATE INDEX ON embeddings USING hnsw (embedding vector_cosine_ops);

Also run supabase_schema_stripe.sql if using the Stripe monetization layer.

---

Environment Variables

Create .env.local from .env.template:

NEXT_PUBLIC_SUPABASE_URL=https://<project>.supabase.co
NEXT_PUBLIC_SUPABASE_ANON_KEY=eyJ...
SUPABASE_SERVICE_ROLE_KEY=eyJ...
GEMINI_API_KEY=AIza...

# Stripe (optional — required for /pricing)
STRIPE_SECRET_KEY=sk_test_...
STRIPE_WEBHOOK_SECRET=whsec_...
STRIPE_PRO_PRICE_ID=price_...
NEXT_PUBLIC_SITE_URL=http://localhost:3000

Note: SUPABASE_SERVICE_ROLE_KEY bypasses Row Level Security. It is used only in server-side actions and API routes — never exposed to the client.

---

Getting Started

npm install
npm run dev

Open http://localhost:3000.

First-time Supabase Setup

1. Create a project at supabase.com
2. Enable the vector extension: Database → Extensions → vector
3. Run supabase_schema.sql in the SQL Editor
4. Copy your project URL and keys into .env.local
5. The documents storage bucket is created automatically on first PDF upload

---

RAG Pipeline

PDF Upload
  │
  ├─ 1. Auto-create storage bucket if missing (supabaseAdmin)
  ├─ 2. Upload file → Supabase Storage → get public URL
  ├─ 3. Parse pages → LangChain WebPDFLoader (page-level metadata)
  ├─ 4. Chunk text → RecursiveCharacterTextSplitter (1000 chars / 200 overlap)
  ├─ 5. Embed chunks → Gemini gemini-embedding-001 (768-dim vectors)
  └─ 6. Store vectors → Supabase pgvector embeddings table

Chat Query
  │
  ├─ 1. Embed user question → Gemini gemini-embedding-001
  ├─ 2. Cosine similarity search → match_embeddings() RPC (top 5 chunks)
  ├─ 3. Filter chunks by documentId, build context with page citations
  └─ 4. Stream response → Gemini gemini-2.5-flash-8b (page-cited Markdown)

---

Chat Page Navigation

The /hub/[id] split-screen chat extracts Page N citations from AI responses and renders them as clickable chips. Clicking a chip updates the PDF iframe URL to #page=N, jumping the viewer to the referenced page. The viewer also auto-jumps on each new AI response.

---

Important Implementation Notes

• Embedding model: text-embedding-004 is not available on this API key; gemini-embedding-001 is used instead (same 768-dim output — no schema change required)
• AI SDK v6: useChat moved from ai/react to @ai-sdk/react — install @ai-sdk/react separately
• pdf-parse version: WebPDFLoader requires pdf-parse@^1 (not v2)
• Server components: Pages using Supabase data are server components; hover effects use CSS classes, not JS event handlers
• Middleware: All app pages (/, /documents, /chat, /hub/*, /api-reference, /settings) are public. Only raw /api/* routes require authentication (Stripe webhook is exempted)

---

Deployment (Vercel)

1. Import repo into Vercel
2. Add all environment variables from .env.template
3. Deploy — vercel.json sets a 60s function timeout to accommodate AI embedding time

---

License

MIT License — Copyright © 2026 Ahmed ELBASHIER. All Rights Reserved.