Update agent.py

agent.py

@@ -8,24 +8,44 @@ class GaiaAgent:
     def __init__(self):
         self.client = OpenAI(api_key=os.getenv("OPENAI_API_KEY"))
         self.instructions = (
-            "You are solving GAIA benchmark questions. "
-            "If given a table, analyze it and extract relevant facts to answer accurately. "
-            "Provide only the final answer. Do not explain."
+            "You are a reasoning assistant solving GAIA benchmark questions. "
+            "If data is provided, analyze it logically and extract the relevant facts. "
+            "Think step by step. Output only the final answer."
         )
         self.api_url = "https://agents-course-unit4-scoring.hf.space"
 
-    def summarize_csv(self, csv_text: str) -> str:
+    def analyze_csv(self, csv_text: str, question: str) -> str:
         try:
             df = pd.read_csv(io.StringIO(csv_text))
-            summary = f"Rows: {len(df)}, Columns: {len(df.columns)}\n"
-            summary += f"Columns: {', '.join(df.columns[:10])}\n"
-            sample_row = df.iloc[0].to_dict()
-            summary += f"Sample row: {sample_row}"
-            return summary
+            question_lower = question.lower()
+
+            if any(k in question_lower for k in ["lowest", "cheapest", "minimum"]):
+                col = self._detect_column(df, ["price", "cost", "amount"])
+                if col:
+                    row = df.sort_values(by=col).iloc[0].to_dict()
+                    return f"Lowest {col}: {row}"
+            elif any(k in question_lower for k in ["highest", "most expensive", "maximum"]):
+                col = self._detect_column(df, ["price", "score", "rating"])
+                if col:
+                    row = df.sort_values(by=col, ascending=False).iloc[0].to_dict()
+                    return f"Highest {col}: {row}"
+            elif "how many" in question_lower:
+                return f"Total rows: {len(df)}"
+            # fallback
+            sample = df.iloc[0].to_dict()
+            return f"Sample row: {sample}"
+
         except Exception as e:
-            return f"[Failed to parse CSV: {e}]"
+            return f"[CSV parsing failed: {e}]"
+
+    def _detect_column(self, df, candidates):
+        for col in df.columns:
+            for name in candidates:
+                if name in col.lower():
+                    return col
+        return None
 
-    def fetch_file_context(self, task_id: str) -> str:
+    def fetch_file_context(self, task_id: str, question: str) -> str:
         try:
             url = f"{self.api_url}/files/{task_id}"
             response = requests.get(url, timeout=10)
@@ -33,26 +53,26 @@ class GaiaAgent:
             content_type = response.headers.get("Content-Type", "")
 
             if "csv" in content_type or url.endswith(".csv"):
-                return self.summarize_csv(response.text)
+                return self.analyze_csv(response.text, question)
             elif "json" in content_type:
-                return f"JSON Sample: {response.text[:1000]}"
+                return f"JSON Preview: {response.text[:1000]}"
             elif "text/plain" in content_type:
                 return f"Text Sample: {response.text[:1000]}"
             elif "pdf" in content_type:
-                return "[PDF detected. OCR not supported in this version.]"
+                return "[PDF detected. OCR not supported.]"
             else:
                 return f"[Unsupported file type: {content_type}]"
 
         except Exception as e:
-            return f"[File fetch error: {e}]"
+            return f"[Error fetching file: {e}]"
 
     def __call__(self, question: str, task_id: str = None) -> str:
-        context = ""
+        file_fact = ""
         if task_id:
-            context = self.fetch_file_context(task_id)
-            context = f"FILE DATA:\n{context}\n"
+            file_fact = self.fetch_file_context(task_id, question)
+            file_fact = f"FILE INSIGHTS:\n{file_fact}\n"
 
-        prompt = f"{self.instructions}\n\n{context}QUESTION: {question}\nANSWER:"
+        prompt = f"{self.instructions}\n\n{file_fact}QUESTION: {question}\nANSWER:"
 
         response = self.client.chat.completions.create(
             model="gpt-4-turbo",