Update app.py

app.py

@@ -2,82 +2,110 @@ import streamlit as st
 import os
 from langchain_huggingface import HuggingFaceEndpoint, ChatHuggingFace
 from langchain_core.prompts import PromptTemplate, FewShotPromptTemplate
-from langchain_core.messages import HumanMessage
+from langchain_core.messages import HumanMessage, SystemMessage
 
 # --- 1. UI Setup ---
-st.set_page_config(page_title="ShotCaller AI", page_icon="🤖")
-st.title("🤖 ShotCaller: Prompt Lab")
-st.markdown("Explore how different prompting techniques change AI reasoning.")
+st.set_page_config(page_title="FlavorFeedback AI", page_icon="🍴")
+st.title("🍴 FlavorFeedback: Prompting Lab")
+st.markdown("""
+This app demonstrates how different **Prompt Engineering** techniques affect AI performance 
+in a Restaurant Management context.
+""")
 
 # --- 2. Model Setup ---
 api_token = os.getenv("HUGGINGFACEHUB_API_TOKEN")
 
 if not api_token:
-    st.error("Missing HUGGINGFACEHUB_API_TOKEN. Add it to your Space Secrets!")
+    st.error("Please add your HUGGINGFACEHUB_API_TOKEN to the Space Secrets.")
     st.stop()
 
-# Switching to Llama-3.1-8B-Instruct (Highly compatible)
-repo_id = "meta-llama/Llama-3.1-8B-Instruct"
+# Using Qwen 2.5 7B - Excellent at reasoning and ungated (no request needed)
+repo_id = "Qwen/Qwen2.5-7B-Instruct"
 
 llm = HuggingFaceEndpoint(
     repo_id=repo_id,
-    task="text-generation", # Base task
+    task="text-generation",
     temperature=0.7,
-    huggingfacehub_api_token=api_token,
-    timeout=300
+    huggingfacehub_api_token=api_token
 )
 
-# This wrapper will now correctly identify Llama as a chat-capable model
 chat_model = ChatHuggingFace(llm=llm)
 
-# --- 3. Sidebar Selection ---
+# --- 3. Sidebar & Logic Selection ---
+st.sidebar.header("Configuration")
 option = st.sidebar.selectbox(
-    "Select Prompting Strategy",
+    "Choose Technique",
     ("Zero-Shot", "Single-Shot", "Few-Shot", "Chain of Thought")
 )
 
-user_query = st.text_input("What's your question?", "Why is the sky blue?")
+# --- 4. Define Defaults for the Use Case ---
+defaults = {
+    "Zero-Shot": "The pasta was okay, but the service was incredibly slow and the waiter forgot our drinks twice.",
+    "Single-Shot": "The staff was so friendly and the steak was cooked to perfection, though the decor felt a bit dated.",
+    "Few-Shot": "The music was way too loud and we couldn't hear each other at the table.",
+    "Chain of Thought": "Issue: Undercooked Salmon. Bill Total: $72. Resolution: Waiter apologized but kept the item on the bill."
+}
 
-# --- 4. Logic for Prompting Techniques ---
-if st.button("Run Inference"):
+user_query = st.text_area("Input Data / Review:", value=defaults[option], height=150)
+
+# --- 5. Execution Logic ---
+if st.button("Generate Response"):
+    system_instruction = "You are a professional Restaurant Operations Assistant."
     formatted_prompt = ""
 
     if option == "Zero-Shot":
-        # Direct question, no guidance
-        formatted_prompt = user_query
+        formatted_prompt = f"Classify the following restaurant review as 'Positive', 'Negative', or 'Neutral':\n\nReview: {user_query}\n\nSentiment:"
 
     elif option == "Single-Shot":
-        # One example to set the tone/format
-        formatted_prompt = f"Example: Input: Hello, Output: Hi there!\nInput: {user_query}\nOutput:"
+        formatted_prompt = (
+            "Extract key ratings from the review.\n\n"
+            "Example:\n"
+            "Input: 'The pizza was amazing, but it was too loud in there.'\n"
+            "Output: Food: 5/5 | Service: N/A | Atmosphere: 2/5\n\n"
+            f"Input: '{user_query}'\n"
+            "Output:"
+        )
 
     elif option == "Few-Shot":
-        # Multiple examples for pattern recognition
-        examples = [
-            {"input": "The movie was great", "output": "Positive"},
-            {"input": "The food was cold", "output": "Negative"},
-            {"input": "It was an okay experience", "output": "Neutral"}
-        ]
-        example_prompt = PromptTemplate(input_variables=["input", "output"], template="Input: {input}\nOutput: {output}")
-        few_shot_p = FewShotPromptTemplate(
-            examples=examples,
-            example_prompt=example_prompt,
-            suffix="Input: {input}\nOutput:",
-            input_variables=["input"]
+        formatted_prompt = (
+            "As the Manager, write a brief response to this feedback.\n\n"
+            "Example 1:\nFeedback: 'Best tacos in town!'\n"
+            "Response: Thank you so much! We're thrilled you enjoyed the tacos.\n\n"
+            "Example 2:\nFeedback: 'Wait time was too long.'\n"
+            "Response: We apologize for the delay. We are working on our speed.\n\n"
+            f"Feedback: {user_query}\n"
+            "Response:"
         )
-        formatted_prompt = few_shot_p.format(input=user_query)
 
     elif option == "Chain of Thought":
-        # Encouraging step-by-step logic
-        formatted_prompt = f"Question: {user_query}\nAnswer: Let's think step by step."
+        system_instruction = "You are a senior restaurant manager who follows strict logic rules."
+        formatted_prompt = (
+            "Rule 1: Complaint must involve Food Quality or Billing.\n"
+            "Rule 2: Total spend must be over $50.\n"
+            "Rule 3: Issue was not resolved on the spot.\n\n"
+            "Determine if this customer gets a 15% discount based on the feedback below.\n"
+            f"Feedback: {user_query}\n\n"
+            "Let's think step-by-step:"
+        )
 
-    # Execute the call
-    with st.spinner("Thinking..."):
+    with st.spinner("Analyzing..."):
         try:
-            # Wrap our formatted string in a HumanMessage for the Chat Model
-            messages = [HumanMessage(content=formatted_prompt)]
+            messages = [
+                SystemMessage(content=system_instruction),
+                HumanMessage(content=formatted_prompt)
+            ]
             response = chat_model.invoke(messages)
             
-            st.subheader(f"Result: {option}")
-            st.write(response.content)
+            st.subheader(f"Results: {option}")
+            st.success(response.content)
+            
+            # Show the "Internal Logic" for the portfolio
+            with st.expander("View the raw prompt sent to AI"):
+                st.code(formatted_prompt)
+                
         except Exception as e:
-            st.error(f"An error occurred: {e}")
+            st.error(f"Error: {e}")
+
+# --- 6. Footer ---
+st.sidebar.markdown("---")
+st.sidebar.info("Built with LangChain & Hugging Face")