Create app.py

app.py

@@ -0,0 +1,81 @@
+# 👮 PromptPolice MVP - Streamlit App with Mistral Backend (JSON Output)
+
+import streamlit as st
+import requests
+import json
+
+HF_API_URL = "https://api-inference.huggingface.co/models/mistralai/Mistral-7B-Instruct-v0.1"
+HF_TOKEN = "HF_PROJECT_TOKEN"
+
+HEADERS = {
+    "Authorization": f"Bearer {HF_TOKEN}",
+    "Content-Type": "application/json"
+}
+
+PROMPT_TEMPLATE = """
+You are a prompt evaluation assistant. Evaluate the following user prompt in JSON format using the structure provided below.
+
+Prompt:
+"""
+{user_prompt}
+"""
+
+Evaluate based on the following criteria:
+- Clarity (1-5)
+- Context (1-5)
+- Specificity (1-5)
+- Intent (1-5)
+Also include a suggestion for improving the prompt.
+
+Respond ONLY in this JSON format:
+{
+  "prompt": "...",
+  "evaluation": {
+    "Clarity": ..., 
+    "Context": ..., 
+    "Specificity": ..., 
+    "Intent": ..., 
+    "suggestion": "..."
+  }
+}
+"""
+
+def evaluate_prompt(user_prompt):
+    payload = {
+        "inputs": PROMPT_TEMPLATE.format(user_prompt=user_prompt),
+        "parameters": {"max_new_tokens": 512, "temperature": 0.7}
+    }
+
+    response = requests.post(HF_API_URL, headers=HEADERS, json=payload)
+    result = response.json()
+
+    # Handle streaming/text output
+    if isinstance(result, list) and "generated_text" in result[0]:
+        raw_text = result[0]["generated_text"]
+        try:
+            # Try to parse the JSON segment only
+            json_part = raw_text[raw_text.index("{"):]
+            return json.loads(json_part)
+        except:
+            return {"error": "Failed to parse model output."}
+    else:
+        return result
+
+# --- Streamlit UI ---
+st.set_page_config(page_title="👮 PromptPolice", layout="centered")
+st.title("👮 PromptPolice: Prompt Evaluator")
+
+user_prompt = st.text_area("Paste your AI prompt here:", height=200)
+
+if st.button("Evaluate Prompt") and user_prompt:
+    with st.spinner("Evaluating your prompt with Mistral..."):
+        evaluation_result = evaluate_prompt(user_prompt)
+
+    st.subheader("Evaluation Result (JSON):")
+    st.code(json.dumps(evaluation_result, indent=2), language='json')
+
+    # Optional: Append to local file for crowd-sourced fine-tuning later
+    if st.button("💾 Save to Dataset"):
+        with open("crowdsourced_prompts.jsonl", "a", encoding="utf-8") as f:
+            f.write(json.dumps(evaluation_result) + "\n")
+        st.success("Prompt evaluation saved!")