docs: add deployment docs and update readme

Author: Ahtisham992

README.md

@@ -4,9 +4,10 @@
   <p><strong>One workspace. Every developer workflow. Zero cloud dependency for AI.</strong></p>
 
   <p>
+    <a href="https://dev-flow-ai-five.vercel.app"><img src="https://img.shields.io/badge/Live_App-Vercel-black?logo=vercel" alt="Live App" /></a>
+    <a href="https://devflow-api-comy.onrender.com"><img src="https://img.shields.io/badge/API-Render-black?logo=render" alt="Live API" /></a>
     <img src="https://img.shields.io/badge/status-Production%20Ready-success" alt="Status" />
     <img src="https://img.shields.io/badge/license-MIT-blue" alt="License" />
-    <img src="https://img.shields.io/badge/platform-Web%20%7C%20iOS%20%7C%20Android-lightgrey" alt="Platforms" />
   </p>
 </div>
 
@@ -28,12 +29,12 @@ DevFlow AI is built as a robust **Monorepo** using the latest enterprise-grade t
 
 | Layer | Technology | Description |
 |---|---|---|
-| **Web App** | Next.js 15, React 19, Tailwind CSS | App Router, Server Components, shadcn/ui components |
+| **Web App** | Next.js 15, React 19 (Hosted on **Vercel**) | App Router, Server Components, shadcn/ui components |
 | **Mobile App** | React Native CLI | Native iOS & Android application sharing the same core backend |
-| **Backend API** | NestJS | Modular architecture, WebSockets (Socket.io) for real-time progress |
-| **Database** | PostgreSQL 16 + pgvector | Relational data + Vector embeddings for semantic AI search |
-| **Cache & Auth** | Redis | JWT Token blacklisting and session management |
-| **AI Runtime** | Ollama | Local execution of Llama 3, Mistral, and Nomic Embeddings |
+| **Backend API** | NestJS (Hosted on **Render**) | Modular architecture, WebSockets for real-time progress |
+| **Database** | PostgreSQL + pgvector (Hosted on **Neon.tech**) | Relational data + Vector embeddings for semantic AI search |
+| **Cache & Auth** | Redis (Hosted on **Upstash**) | JWT Token blacklisting and session management |
+| **AI Runtime** | Ollama (Hosted on **Oracle Cloud VM**) | Remote execution of Llama 3 and Nomic Embeddings |
 | **ORM** | Prisma 7 | Type-safe database interactions and automated migrations |
 | **State** | Zustand + React Query | Global UI state management and server-data caching |
 
@@ -129,6 +130,7 @@ For an exhaustive, deep-dive look into how this system was built, please refer t
 - `DevFlowAI-Documentation_1.md`: Weeks 1-3 (Foundation & Auth)
 - `DevFlowAI-Documentation_2.md`: Weeks 4-10 (RAG, WebSockets, GitHub Indexing)
 - `DevFlowAI-Documentation_3.md`: Weeks 11-12 (Mobile App & Polish)
+- `DevFlowAI-Documentation_4.md`: Weeks 13-14 (Production Deployment, Neon, Render, Oracle Cloud, E2E & Load Testing)
 
 ---
 

documents/DevFlowAI-Documentation_4.md

@@ -0,0 +1,55 @@
+# DevFlow AI: Production Deployment & Verification
+
+## 1. Cloud Infrastructure & Architecture
+
+Moving from a local Docker Compose setup to a production-ready cloud architecture required distributing our services across multiple specialized platforms to balance cost, performance, and scalability. The final production architecture is 100% free-tier compliant while maintaining high availability.
+
+### 1.1 Web Frontend (Vercel)
+- **Host**: Vercel
+- **Live URL**: `https://dev-flow-ai-five.vercel.app`
+- **Why**: Vercel provides zero-config Next.js deployments, global CDN caching, and automatic CI/CD from GitHub. 
+- **Configuration**: The `NEXT_PUBLIC_API_URL` environment variable was set to point to the production Render backend instead of localhost.
+
+### 1.2 Backend API (Render)
+- **Host**: Render (Web Service)
+- **Live URL**: `https://devflow-api-comy.onrender.com`
+- **Why**: Render offers native Node.js/NestJS support, easy environment variable management, and automatic HTTPS.
+- **Configuration**: Configured CORS to whitelist the Vercel frontend domain, preventing unauthorized clients from accessing the API.
+
+### 1.3 PostgreSQL Database (Neon)
+- **Host**: Neon.tech
+- **Why**: Serverless Postgres that scales to zero and supports the `pgvector` extension out of the box, which is critical for our RAG (Retrieval-Augmented Generation) embeddings.
+- **Configuration**: The `DATABASE_URL` was securely stored in Render. We executed `npx prisma migrate deploy` and `npx prisma db seed` during the build step.
+
+### 1.4 Redis Cache (Upstash)
+- **Host**: Upstash
+- **Why**: Serverless Redis with a generous free tier. It provides the exact same API as local Redis, making it a drop-in replacement.
+- **Usage**: Used for BullMQ background job processing, caching GitHub repositories, and JWT session blacklisting.
+
+### 1.5 AI Runtime (Oracle Cloud Infrastructure)
+- **Host**: Oracle Cloud "Always Free" ARM VM
+- **IP**: `http://152.67.112.59:11434`
+- **Why**: Ollama requires significant RAM to run models like LLaMA 3. Oracle Cloud provides a 24GB RAM / 4 OCPU ARM instance entirely for free, which is perfect for hosting an LLM API.
+- **Configuration**: Configured `OLLAMA_HOST=0.0.0.0` and opened port `11434` in the Oracle Cloud VCN ingress rules. The backend connects to this via the `OLLAMA_URL` environment variable.
+
+---
+
+## 2. Production Verification & Testing
+
+To ensure the system works reliably under real-world conditions, we implemented two rigorous testing suites.
+
+### 2.1 End-to-End (E2E) Testing with Playwright
+We implemented automated browser tests to verify critical user flows in the live production environment.
+- **Framework**: Playwright
+- **Coverage**: User Registration, Login, Workspace Creation, Project Setup.
+- **Execution**: The tests were configured to dynamically use `PLAYWRIGHT_TEST_BASE_URL=https://dev-flow-ai-five.vercel.app`.
+- **Result**: The test suite successfully completed all end-to-end flows in **15.8 seconds** with zero errors.
+
+### 2.2 API Load Testing with k6
+To verify that our free-tier architecture (specifically Neon and Render) wouldn't collapse under a sudden influx of traffic, we used `k6` to stress-test the backend.
+- **Framework**: k6 by Grafana
+- **Scenario**: 50 concurrent Virtual Users (VUs) constantly hitting the `/` endpoint and the `/auth/login` endpoint for 60 seconds.
+- **Result**: 
+  - Over 1,800 requests were processed.
+  - The NestJS `ThrottlerModule` (Rate Limiter) successfully intercepted the flood. It allowed exactly 100 requests per IP and returned `429 Too Many Requests` for the rest, preventing database connection exhaustion.
+  - 95th percentile latency (p95) for successful requests remained under 800ms.