Create Logic for Lambda Function

Create Logic for Lambda Function

Implementation Steps

1. Access Lambda Service

• Go to AWS Management Console → search for Lambda.

• First, go to the Layers section to create a layer so the Lambda function has the library to connect to PostgreSQL (psycopg2).

• Access the GitHub link to download the appropriate Python version, here we use psycopg2-3.11: https://github.com/jkehler/awslambda-psycopg2

• After downloading, put the entire psycopg2-3.11 folder into a folder named (python), then zip that folder and name it (postgres-layer-3.11).

• Click Create layer, name the layer, upload the postgres-layer-3.11 zip file → Create.

2. Create Lambda

• Functions → Create function.

• Name the Lambda function and select runtime Python 3.11.

• In Additional configurations, check VPC, select the created VPC, select 1 private subnet for Subnet, and select Lambda-SG for Security group.

• After the Lambda function is created, we will go to the layer section to add the created layer.

• Select Custom layers, choose the created layer, select version 1 → Add.

• Go to the Configuration tab, in General configuration, increase the timeout to 1 minute.

• In Permissions, add AmazonBedrockFullAccess and AWSLambdaVPCAccessExecutionRole.

• In Environment variables, add the following variables: DB_HOST, DB_NAME, DB_USER, DB_PASS (Enter your RDS information).

• After configuring, paste this Python code into the Lambda function and click Deploy.

  import json
  import boto3
  import psycopg2
  import os
  
  # --- CONFIGURATION ---
  DB_HOST = os.environ.get('DB_HOST')
  DB_NAME = os.environ.get('DB_NAME')
  DB_USER = os.environ.get('DB_USER')
  DB_PASS = os.environ.get('DB_PASS')
  
  # Use ap-southeast-1 region
  bedrock = boto3.client(service_name='bedrock-runtime', region_name='ap-southeast-1')
  
  # --- 1. EMBEDDING FUNCTION (COHERE) ---
  def get_embedding(text):
      try:
          body = json.dumps({
              "texts": [text],
              "input_type": "search_query",
              "truncate": "END"
          })
          response = bedrock.invoke_model(
              body=body,
              modelId="cohere.embed-multilingual-v3",
              accept="application/json", contentType="application/json"
          )
          return json.loads(response['body'].read())['embeddings'][0]
      except Exception as e:
          print(f"Embed Error: {e}")
          return None
  
  # --- 2. SAFE METADATA HANDLING FUNCTION ---
  def format_product_info(metadata):
      """
      This function helps normalize data even if the CSV file is missing columns
      """
      if not metadata: return None
  
      # Get name (Prioritize common keys)
      name = metadata.get('name') or metadata.get('product_name') or metadata.get('title') or "Unnamed Product"
  
      # Get price (If not present, leave empty or 'Contact')
      price = metadata.get('price') or metadata.get('gia') or metadata.get('display_price')
      price_str = f"- Price: {price}" if price else ""
  
      # Get type (if available from smart import script)
      category = metadata.get('type') or "Product"
  
      # Create description string for AI to read
      # Example: "Bird Species: Parrot. - Price: 500k"
      ai_context = f"Category/Product: {name} ({category}) {price_str}"
  
      # Create object for Frontend display (Product Card)
      frontend_card = {
          "id": metadata.get('id') or metadata.get('product_id'),
          "name": name,
          "price": price if price else "Contact", # Frontend will show "Contact" if no price
          "image": metadata.get('image_url') or metadata.get('link_anh') or "", # Image can be empty
          "type": category # To let frontend know if this is a bird or a toy
      }
  
      return ai_context, frontend_card
  
  # --- 3. MAIN HANDLER ---
  def lambda_handler(event, context):
      print("Event:", event)
  
      try:
          # Parse Input
          if 'body' in event:
              try:
                  body_data = json.loads(event['body']) if isinstance(event['body'], str) else event['body']
              except: body_data = {}
          else: body_data = event
  
          user_question = body_data.get('question', '')
          if not user_question: return {'statusCode': 400, 'body': json.dumps('Missing question')}
  
          # A. Create Vector
          q_vector = get_embedding(user_question)
          if not q_vector: return {'statusCode': 500, 'body': json.dumps('Vector creation error')}
  
          # B. Search DB
          conn = psycopg2.connect(host=DB_HOST, database=DB_NAME, user=DB_USER, password=DB_PASS)
          cur = conn.cursor()
  
          # Retrieve metadata for processing
          sql = """
              SELECT content, metadata
              FROM knowledge_base
              ORDER BY embedding <=> %s
              LIMIT 3
          """
          cur.execute(sql, (json.dumps(q_vector),))
          results = cur.fetchall()
          cur.close(); conn.close()
  
          # C. Process Results (MOST IMPORTANT)
          ai_contexts = []
          frontend_products = []
  
          for row in results:
              raw_content = row[0] # Original content at import
              raw_metadata = row[1] # JSON metadata
  
              # Call normalization function
              ai_text, card_data = format_product_info(raw_metadata)
  
              if ai_text:
                  # Merge original content + identification content to be sure
                  ai_contexts.append(f"{ai_text}. Details: {raw_content}")
                  frontend_products.append(card_data)
  
          # If nothing is found
          if not ai_contexts:
              final_answer = "Sorry, the shop currently cannot find matching information in the database."
          else:
              # D. Send to LLM (Claude 3 Haiku)
              context_str = "\n---\n".join(ai_contexts)
  
              system_prompt = (
                  "You are a consultant for the Bird Shop, named 'Smart Parrot'. "
                  "Style: Friendly, Polite but Concise.\n\n"
  
                  "HANDLING INSTRUCTIONS (PRIORITIZE IN ORDER):\n"
                  "1. SOCIAL INTERACTION: If the customer just says hello (Hello, Hi...) or thanks: "
                  "-> Greeting back warmly and ask what they are looking for. DO NOT list products unless asked.\n"
  
                  "2. PRODUCT CONSULTATION: When customer asks about goods:\n"
                  "-> Answer concisely: Product Name + Price + Stock status (if any).\n"
                  "-> Do not describe in flowery detail unless asked specifically 'how is it'.\n"
                  "-> If information is not in Context, say 'Sorry, the shop doesn't have this item yet'.\n"
              )
  
              user_msg = f"Reference Information:\n{context_str}\n\nQuestion: {user_question}"
  
              # Call Claude 3 Haiku
              claude_body = json.dumps({
                  "anthropic_version": "bedrock-2023-05-31",
                  "max_tokens": 300,
                  "temperature": 0.1,
                  "system": system_prompt,
                  "messages": [{"role": "user", "content": user_msg}]
              })
  
              try:
                  model_res = bedrock.invoke_model(
                      body=claude_body,
                      modelId="anthropic.claude-3-haiku-20240307-v1:0"
                  )
                  res_json = json.loads(model_res['body'].read())
                  final_answer = res_json['content'][0]['text']
              except Exception as e:
                  print(f"LLM Call Error: {e}")
                  final_answer = "Sorry, the AI system is busy, please try again later."
  
          # E. Return Result
          return {
              'statusCode': 200,
              'headers': {
                  'Access-Control-Allow-Origin': '*',
                  'Access-Control-Allow-Methods': 'POST',
                  'Content-Type': 'application/json'
              },
              'body': json.dumps({
                  "answer": final_answer,
                  "products": frontend_products # Frontend uses this to draw UI
              })
          }
  
      except Exception as e:
          print(f"System Error: {e}")
          return {'statusCode': 500, 'body': json.dumps(str(e))}
  

3. Integrate into Team’s API Gateway

• Access API Gateway Service.

• Select the API created by the Backend.

• Select Create resource.

• Resource name chatbot and check CORS.

• Select the chatbot resource and click Create method.

• Method type select POST, check Lambda proxy integration.

• Select the created Lambda Function (or VPC based on original context), then click Create method.

• After configuration is complete, click Deploy API.