add agent logic

app.py

@@ -10,14 +10,92 @@ DEFAULT_API_URL = "https://agents-course-unit4-scoring.hf.space"
 
 # --- Basic Agent Definition ---
 # ----- THIS IS WERE YOU CAN BUILD WHAT YOU WANT ------
+# --- Enhanced Agent (replace the old BasicAgent with this) ---
+import re
+import json
+import textwrap
+
 class BasicAgent:
+    """
+    Calls a small LLM on the Hugging Face Inference API and post-processes the output
+    to return a concise, exact-match-friendly answer.
+    Reads HF API token from the HF Space secret HF_TOKEN.
+    """
     def __init__(self):
-        print("BasicAgent initialized.")
+        self.model_url = (
+            "https://api-inference.huggingface.co/models/HuggingFaceH4/zephyr-7b-beta"
+            # You can swap to another text-generation-inference model on the Hub.
+        )
+        self.headers = {
+            "Authorization": f"Bearer {os.getenv('HF_TOKEN', '')}",
+            "Content-Type": "application/json",
+        }
+        if not os.getenv("HF_TOKEN"):
+            print("⚠️  HF_TOKEN not set — requests to the Inference API will fail.")
+        print("🚀 Enhanced BasicAgent initialized.")
+
+    def _clean(self, raw: str) -> str:
+        """
+        Post-process the model output to a short, exact value:
+        - Keep only the final line after common prefixes.
+        - Strip quotes/backticks and trailing punctuation.
+        """
+        txt = raw.strip()
+
+        # If the model echoed the prompt, take the last line
+        lines = [l.strip() for l in txt.splitlines() if l.strip()]
+        if lines:
+            txt = lines[-1]
+
+        # Remove common prefixes
+        txt = re.sub(r"^(final answer|answer|prediction)\s*[:\-]\s*", "", txt, flags=re.I)
+
+        # Strip code fences/quotes and trailing punctuation
+        txt = txt.strip("`'\" \t\n\r")
+        txt = re.sub(r"[ \t]*[.;,:-]+$", "", txt)
+
+        # Keep it short (exact match tasks usually want a short string/number)
+        return txt[:200]
+
     def __call__(self, question: str) -> str:
-        print(f"Agent received question (first 50 chars): {question[:50]}...")
-        fixed_answer = "This is a default answer."
-        print(f"Agent returning fixed answer: {fixed_answer}")
-        return fixed_answer
+        print(f"🧠 Agent received question: {question[:120]}...")
+        # A very simple, direct prompt to reduce rambling
+        prompt = textwrap.dedent(f"""
+        You must answer the question with a single, concise value
+        (number, word, date, short phrase) and nothing else.
+
+        Question: {question}
+        Final answer:
+        """).strip()
+
+        payload = {
+            "inputs": prompt,
+            "parameters": {
+                "max_new_tokens": 96,
+                "temperature": 0.2,
+                "return_full_text": False,
+            },
+            "options": {"wait_for_model": True},
+        }
+
+        try:
+            r = requests.post(self.model_url, headers=self.headers, data=json.dumps(payload), timeout=60)
+            r.raise_for_status()
+            data = r.json()
+            # Inference API (text-generation) usually returns a list with "generated_text"
+            if isinstance(data, list) and data and "generated_text" in data[0]:
+                generated = data[0]["generated_text"]
+            else:
+                # Fallback: stringify whatever came back
+                generated = str(data)
+            answer = self._clean(generated)
+            print(f"✅ Cleaned answer: {answer}")
+            # Ensure we never submit an empty string (server rejects)
+            return answer if answer else "N/A"
+        except Exception as e:
+            print(f"❌ Error generating answer: {e}")
+            return "N/A"
+
 
 def run_and_submit_all( profile: gr.OAuthProfile | None):
     """