app.py

import re
import streamlit as st
from together import Together

# Initialize the page
st.set_page_config(page_title="AI Prompt Evaluator", page_icon="🧠", layout="wide")
st.title("AI Prompt Evaluator")

def evaluate_prompt(prompt):
    """Evaluates a prompt based on key principles of good prompt engineering."""
    
    criteria = {
        "clarity": bool(re.search(r"\b(who|what|where|when|why|how)\b", prompt.lower())),
        "specificity": len(prompt.split()) > 5,
        "context": bool(re.search(r"\b(for example|such as|like|including)\b", prompt.lower())),
        "output_format": bool(re.search(r"\b(list|table|bullets|code|summary)\b", prompt.lower())),
        "constraints": bool(re.search(r"\b(limit|max|min|exactly|within)\b", prompt.lower()))
    }
    
    score = sum(criteria.values()) * 2  # Scale to 10 points
    
    suggestions = []
    if not criteria["clarity"]:
        suggestions.append("Make the prompt clearer by specifying exactly what you need.")
    if not criteria["specificity"]:
        suggestions.append("Make the prompt more detailed and specific.")
    if not criteria["context"]:
        suggestions.append("Add some background information to improve relevance.")
    if not criteria["output_format"]:
        suggestions.append("Specify how you want the output to be structured (e.g., list, code).")
    if not criteria["constraints"]:
        suggestions.append("Define any limits or conditions (e.g., word count, time frame).")
    
    return score, suggestions, criteria

def get_ai_response(prompt, model):
    """Get a response from the Together API"""
    try:
        client = Together(api_key=st.secrets["together_api_key"])
        response = client.chat.completions.create(
            model=model,
            messages=[{"role": "user", "content": prompt}],
            max_tokens=500
        )
        return response.choices[0].message.content
    except Exception as e:
        return f"Error: {str(e)}"

# Sidebar for configuration
st.sidebar.header("Configuration")
model_options = {
    "Mistral 7B": "mistralai/Mistral-7B-Instruct-v0.3",
    "Llama 3 8B": "meta-llama/Llama-3-8B-Instruct",
    "Qwen 72B": "Qwen/Qwen-72B-Chat"
}
selected_model = st.sidebar.selectbox("Select AI Model", list(model_options.keys()))

# Main input area
prompt_input = st.text_area("Enter your prompt:", height=150)

col1, col2 = st.columns(2)
with col1:
    if st.button("Evaluate Prompt", type="primary"):
        if not prompt_input.strip():
            st.error("Please enter a prompt to evaluate.")
        else:
            # Store API key in secrets
            if "together_api_key" not in st.secrets:
                st.secrets["together_api_key"] = "4db152889da5afebdba262f90e4cdcf12976ee8b48d9135c2bb86ef9b0d12bdd"
            
            score, suggestions, criteria = evaluate_prompt(prompt_input)
            
            # Display evaluation results
            st.subheader("Prompt Evaluation Results")
            
            # Score with color-coded meter
            if score >= 8:
                st.success(f"Score: {score}/10 - Excellent!")
            elif score >= 6:
                st.info(f"Score: {score}/10 - Good")
            else:
                st.warning(f"Score: {score}/10 - Needs Improvement")
            
            # Criteria checklist
            st.markdown("#### Criteria Checklist")
            for criterion, passed in criteria.items():
                icon = "✅" if passed else "❌"
                st.markdown(f"{icon} **{criterion.capitalize()}**")
            
            # Improvement suggestions
            if suggestions:
                st.markdown("#### Suggestions to improve your prompt:")
                for tip in suggestions:
                    st.markdown(f"- {tip}")
            else:
                st.success("Your prompt is well-structured!")
            
            # Get AI response if score is high enough
            if score >= 4:  # Only get response if score is decent
                with st.spinner("Getting AI response..."):
                    ai_response = get_ai_response(prompt_input, model_options[selected_model])
                    
                st.subheader("AI Response Preview")
                st.markdown(ai_response)
            else:
                st.info("Improve your prompt score to see an AI response preview.")

with col2:
    st.subheader("Prompt Engineering Tips")
    st.markdown("""
    ### How to craft effective prompts:
    
    1. **Be clear and specific** - Clearly state what you want.
    2. **Provide context** - Give background information.
    3. **Specify output format** - Ask for lists, code, or summaries.
    4. **Set constraints** - Define limits like word count.
    5. **Use examples** - Show examples of desired output.
    
    ### Examples of good prompts:
    
    - "Create a 5-item bulleted list of healthy breakfast ideas including nutritional benefits."
    - "Write a Python function that sorts a list of integers using the bubble sort algorithm."
    - "Summarize the key benefits of exercise in exactly 100 words for a health newsletter."
    """)

# Add explanation of the evaluation criteria
st.markdown("---")
with st.expander("Understanding the Evaluation Criteria"):
    st.markdown("""
    - **Clarity**: Does the prompt include clear question words (who, what, where, when, why, how)?
    - **Specificity**: Is the prompt detailed enough (more than 5 words)?
    - **Context**: Does the prompt provide examples or background information?
    - **Output Format**: Does the prompt specify the desired format (list, table, code, etc.)?
    - **Constraints**: Does the prompt include specific limitations or requirements?
    """)

# Footer
st.markdown("---")
st.markdown("#### About")
st.markdown("This tool helps you create better prompts for AI systems by evaluating them against best practices in prompt engineering.")