03 Create a Chat GPT-like application (RAG-based AI app) using Azure Services - Details Design

03 Create a ChatGPT-like application (RAG-based AI app) using Azure Services

---

🧩 1. Clean Azure Architecture (ChatGPT-style App)

6

🧠 How to read this:

Flow:

1. User → Frontend (UI)
2. Frontend → Backend API
3. Backend → Azure AI Search (RAG)
4. Backend → Azure OpenAI
5. Response → User

---

🏗️ Service Mapping

Layer	Azure Service
Frontend	Azure Static Web Apps
Backend	Azure Functions / App Service
AI Model	Azure OpenAI Service
RAG	Azure AI Search
Storage	Blob Storage + Cosmos DB
Security	Microsoft Entra ID + Key Vault
Monitoring	Application Insights

---

🛠️ 2. Step-by-Step Build Guide (Hands-On)

🔹 Step 1: Create Azure Resources

Go to Microsoft Azure Portal and create:

• Azure OpenAI resource
• Azure AI Search
• Storage Account (Blob)
• Azure Functions
• Static Web App

---

🔹 Step 2: Upload Your Data (RAG Setup)

• Upload PDFs to Blob Storage
• Extract + chunk text (Python script)
• Generate embeddings using Azure OpenAI Service
• Store in Azure AI Search

---

🔹 Step 3: Backend API (Core Logic)

Example (Python - Azure Function):

def chat(req):
    user_query = req.json["message"]

    # 1. Search relevant docs
    docs = search_index(user_query)

    # 2. Build prompt
    prompt = f"Answer using context: {docs}\nQuestion: {user_query}"

    # 3. Call OpenAI
    response = openai.chat.completions.create(
        model="gpt-4",
        messages=[{"role": "user", "content": prompt}]
    )

    return response.choices[0].message.content

---

🔹 Step 4: Frontend (Simple UI)

Use React:

• Input box (like ChatGPT)
• Call backend API
• Display response

Deploy via:

• Azure Static Web Apps

---

🔹 Step 5: Add Authentication

• Use Microsoft Entra ID
• Enable login (Google / Microsoft)

---

🔹 Step 6: Store Chat History

• Use Cosmos DB:
• user_id
• message
• response

---

🔹 Step 7: Monitor

• Enable Azure Monitor
• Add logs + alerts

---

💰 3. Cost Estimate (Realistic Monthly)

Here’s a starter-level production estimate 👇

---

🧠 AI (Azure OpenAI)

• GPT-4 / GPT-4o usage
  👉 ~$10 – $100/month (depends on usage)

---

🔎 Azure AI Search

• Basic tier
  👉 ~$75/month

---

💾 Storage

• Blob Storage
  👉 ~$5 – $20/month

---

⚙️ Backend

• Azure Functions (consumption plan)
  👉 ~$0 – $20/month

---

🌐 Frontend

• Static Web Apps
  👉 Free tier available

---

🧾 Database

• Cosmos DB
  👉 ~$25 – $100/month

---

📊 Monitoring

• Application Insights
  👉 ~$5 – $20/month

---

💵 Total Estimate

Scale	Monthly Cost
🧪 Small (testing)	$20 – $50
🚀 Startup MVP	$100 – $300
🏢 Production	$500+

---

🧠 Final Insight (Very Important)

👉 The most expensive part is NOT Azure itself—it’s AI tokens

So optimize:

• Prompt size
• Retrieval quality (RAG)
• Caching responses