How arhanansari/chat-to-pdf works
Project type and purpose
• This is a full-stack SaaS web application (“Chat to PDF”) built with Next.js and React, integrating AI (OpenAI & Langchain), vector DB (Pinecone), file storage (Firebase), authentication (Clerk), and payment (Stripe). Users sign up, upload PDFs, chat with their documents, and manage subscriptions.
File structure
• Top-level:
• app/ (Next.js 13+ App Router) with pages: landing (app/page.tsx), sign-in/up, dashboard, upload, files, upgrade, webhook
• actions/ (server actions for Stripe, Firestore, embeddings, document delete)
• components/ & components/ui/ (React UI components & design system)
• lib/ (wrappers for OpenAI/Langchain, Pinecone, Stripe, utils)
• hooks/ (client hooks: useUpload, useSubscription)
• firebase.ts & firebaseAdmin.ts (client & admin SDK init)
• middleware.ts (route protection via Clerk)
• next.config.mjs, tailwind.config.ts, etc.
README for additional insights
• Describes a 5-day challenge demo app: build landing, auth with Clerk, upload with React-Dropzone, Firebase storage & Firestore, PDF viewer, Pinecone + Langchain + OpenAI chat, Stripe checkout & webhooks, role-based access (free / pro), file delete/download, deployments to Vercel.
a. Main components
• Client (Browser/React UI)
• Next.js App (Server + Edge Functions)
• Authentication Service (Clerk)
• File Storage & Database (Firebase Storage & Firestore)
• Vector DB & Embedding Service (Pinecone + Langchain + OpenAI API)
• Payment Service (Stripe + Webhook)
• External CDN/Deployment (Vercel / Netlify badge)
b. Relationships & interactions
• User ←→ Next.js frontend (React on client, server-side rendering/page routing)
• Next.js ←→ Clerk for sign-in/up, JWT session
• Next.js Server Actions / API Routes:
• upload PDF → Firebase Storage & store metadata in Firestore
• generateEmbeddings → OpenAI embeddings → Pinecone upsert
• askQuestion → Query Pinecone → Langchain chain → OpenAI → return chat response
• createCheckoutSession / createStripePortal → Stripe API → return session URL
• Stripe Webhook endpoint (app/webhook/route.ts) → Listen for events → update Firestore subscription docs
• deleteDocument → remove from Firestore & Pinecone & Firebase Storage
• Client Hooks consume:
• useUpload (calls upload action)
• useSubscription (polls Firestore for subscription status)
• Components render PDF (PdfView), chat (Chat, ChatMessage), documents list, config form, upgrade flows.
c. Architectural patterns & design principles
• Jamstack / Serverless functions (Next.js App Router server actions)
• Separation of concerns: UI components, server actions, data layer (lib/ + hooks/)
• External managed services for heavy lifting (auth, AI, vector DB, payments, storage)
• Single-page app feel with SSR & ISR where needed
d. Key technologies & libraries
• Next.js 13 (App Router, React Server Components)
• React & Tailwind CSS (shadcn/ui primitives)
• Clerk (authentication & middleware)
• Firebase Storage & Firestore (file storage & metadata)
• Pinecone + Langchain + OpenAI (vector search & chat)
• Stripe (checkout, billing portal, webhooks)
• Vercel (deployment)
• Frontend: Next.js pages & React components communicate via fetch or server actions.
• Auth: Clerk middleware protects dashboard routes; tokens validated server-side.
• File upload: React-Dropzone → Next.js action → Firebase Storage → Firestore metadata.
• Embeddings: upon file upload, server action calls OpenAI embeddings → Pinecone client in lib/pinecone.ts.
• Chat: frontend sends question → askQuestion action → Pinecone query → Langchain chain → OpenAI → return answer.
• Billing: createCheckoutSession → Stripe checkout → webhook endpoint updates Firestore subscription status → UI hook reflects plan.
• Document management: list & delete documents via server actions that coordinate Firebase + Pinecone.