Amending prompt - adding more examples for few shot learning

app.py

@@ -70,7 +70,7 @@ class BasicAgent:
         1. **Thought:** Briefly outline your reasoning step.  
         2. **Reflect:** Check “Did I use all observations? Did my tool call succeed?”  
         3. **Action:** Either call a tool (with arguments) or prepare your final answer.  
-        4. **Final Answer:** Provide only the bare result (no labels, no extra text, no thoughts, no reflection, no "Final Answer" string in the result). For question that contain phrases like `what is the number` or 
+        4. **Final Answer:** Provide only the bare result (no labels, no extra text, no actions, no thoughts, no reflection, no "Final Answer" string in the result). For question that contain phrases like `what is the number` or 
         `what is the highest number` return just the number, e.g., 2.
         
         **Answer Format Rules**  
@@ -96,7 +96,19 @@ class BasicAgent:
         Thought: Alphabetical descending: cherry, banana, apple.  
         Reflect: Order and formatting confirmed.  
         Final Answer: cherry, banana, apple
-        
+
+        **Example 4**
+        Q: The attached csv file contains the amount of impressions for an ad campaign. What were the total amount of clicks crevenue that occurred after 2024-01-01? Express your answer in EUR with two decimal places.
+        Thought: Calculate the total amount of revenue for clicks across all dates after 2024-01-01.
+        Reflect: I have all the necessary data from the csv file.
+        Action: Multiple clicks amount by revenue per click for each row after 2024-01-01 and then sum these values.
+        Final Answer: 283934.00
+
+        **Example 5**
+        Q: What is the number of the most performant desktop processor model from Ryzen 1000 series?
+        Thought: The number of the most performant desktop processor model from Ryzen 1000 series is 1800X.
+        Reflect: I know the answer, displaying only the model number without anything else.
+        Final Answer: 1800X
         ---
         
         Now answer the next question following this chain-of-thought + reflection pattern, and output **only** the `Final Answer` in the required format.
@@ -178,18 +190,18 @@ class BasicAgent:
         with open(path, "rb") as f:
             b64 = b64encode(f.read()).decode()
 
-        file_msg = {
-            "type": "file",
-            "file": {
-                "data": b64,
-                "filename": os.path.basename(path)
-            }
-        }
-    
+        ext = os.path.splitext(path)[1].lower().lstrip(".")
+        mime = f"image/{'jpeg' if ext in ('jpg','jpeg') else 'png'}"
+        
+        # 2) Build the multimodal message
         msg = HumanMessage(content=[
-            {"type":"text", "text": question},
-            file_msg
+            {"type": "text", "text": question},
+            {
+              "type": "image_url",
+              "image_url": {"url": f"data:{mime};base64,{b64}"}
+            }
         ])
+        
         response = model.invoke([SystemMessage(content="Analyze the image."), msg])
         result = response.content
         print("Result:", result)