Create chatbot_handler.py

chatbot_handler.py

@@ -0,0 +1,116 @@
+# chatbot_handler.py
+import logging
+import json
+import aiohttp # Using asynchronous aiohttp
+import os
+
+# Ensure GROQ_API_KEY is set in your environment variables
+api_key = os.getenv('GEMINI_API_KEY')
+
+def format_history_for_gemini(gradio_chat_history: list) -> list:
+    """
+    Converts Gradio chat history (list of dicts with 'role' and 'content')
+    to Gemini API's 'contents' format.
+    Gemini expects roles 'user' and 'model'.
+    It also filters out system messages if any, as Gemini handles system prompts differently.
+    """
+    gemini_contents = []
+    for msg in gradio_chat_history:
+        role = "user" if msg["role"] == "user" else "model" # Gemini uses 'model' for assistant
+        # Ensure content is a string, skip if not (e.g. if a gr.Plot was accidentally in history)
+        if isinstance(msg.get("content"), str):
+            gemini_contents.append({"role": role, "parts": [{"text": msg["content"]}]})
+        else:
+            logging.warning(f"Skipping non-string content in chat history for Gemini: {msg.get('content')}")
+    return gemini_contents
+
+async def generate_llm_response(user_message: str, plot_id: str, plot_label: str, chat_history_for_plot: list):
+    """
+    Generates a response from the LLM using Gemini API.
+    Args:
+        user_message (str): The latest message from the user.
+        plot_id (str): The ID of the plot being discussed.
+        plot_label (str): The label of the plot being discussed.
+        chat_history_for_plot (list): The current conversation history for this plot.
+                                      This list already includes the latest user_message.
+    Returns:
+        str: The LLM's response text.
+    """
+    logging.info(f"Generating LLM response for plot_id: {plot_id} ('{plot_label}'). User message: '{user_message}'")
+
+    # The chat_history_for_plot already contains the full conversation including the latest user message.
+    # The initial system-like prompt is the first message from the assistant in the history.
+    gemini_formatted_history = format_history_for_gemini(chat_history_for_plot)
+
+    if not gemini_formatted_history:
+        logging.error("Cannot generate LLM response: Formatted history is empty.")
+        return "I'm sorry, there was an issue processing the conversation history."
+
+    # Construct payload for Gemini API
+    payload = {
+        "contents": gemini_formatted_history,
+        "generationConfig": { # Optional: configure generation parameters
+            "temperature": 0.7,
+            "topK": 1,
+            "topP": 1,
+            "maxOutputTokens": 2048,
+        }
+    }
+    
+
+    # Using gemini-2.0-flash as per instructions
+    apiUrl = f"https://generativelanguage.googleapis.com/v1beta/models/gemini-2.5-flash-preview-05-20:generateContent?key={api_key}"
+
+    async with aiohttp.ClientSession() as session:
+        try:
+            async with session.post(apiUrl, headers={'Content-Type': 'application/json'}, json=payload, timeout=45) as resp:
+                response_json = await resp.json()
+                logging.debug(f"LLM API Raw Response for '{plot_label}': {json.dumps(response_json, indent=2)}")
+
+                if resp.status != 200:
+                    error_detail = response_json.get('error', {}).get('message', 'Unknown API error')
+                    # Check for specific content policy block from Gemini, even on error status
+                    if response_json.get("promptFeedback") and response_json["promptFeedback"].get("blockReason"):
+                         reason = response_json["promptFeedback"]["blockReason"]
+                         safety_ratings_info = [f"{rating['category']}: {rating['probability']}" for rating in response_json['promptFeedback'].get('safetyRatings', [])]
+                         details = f" Safety Ratings: {', '.join(safety_ratings_info)}" if safety_ratings_info else ""
+                         logging.warning(f"Content blocked by API (Status {resp.status}) for '{plot_label}'. Reason: {reason}.{details}")
+                         return f"I'm sorry, I can't respond to that due to content policy: {reason}."
+                    logging.error(f"LLM API Error (Status {resp.status}) for '{plot_label}': {error_detail}")
+                    return f"Sorry, the AI model returned an error (Status {resp.status}). Please try again later."
+
+                if response_json.get("candidates") and \
+                   response_json["candidates"][0].get("content") and \
+                   response_json["candidates"][0]["content"].get("parts") and \
+                   response_json["candidates"][0]["content"]["parts"][0].get("text"):
+                    response_text = response_json["candidates"][0]["content"]["parts"][0]["text"]
+                    logging.info(f"LLM generated response for '{plot_label}': {response_text[:100]}...")
+                    return response_text
+                # Check for promptFeedback even on 200 if candidates are missing (e.g. blocked content)
+                elif response_json.get("promptFeedback") and response_json["promptFeedback"].get("blockReason"):
+                    reason = response_json["promptFeedback"]["blockReason"]
+                    safety_ratings_info = [f"{rating['category']}: {rating['probability']}" for rating in response_json['promptFeedback'].get('safetyRatings', [])]
+                    details = f" Safety Ratings: {', '.join(safety_ratings_info)}" if safety_ratings_info else ""
+                    logging.warning(f"Content blocked by API (Status 200 but no candidate) for '{plot_label}'. Reason: {reason}.{details}")
+                    return f"I'm sorry, your request was processed but could not be answered due to content policy: {reason}."
+                else:
+                    logging.error(f"Unexpected LLM API response structure for '{plot_label}': {response_json}")
+                    return "Sorry, I received an unexpected or empty response from the AI model."
+
+        except aiohttp.ClientTimeout:
+            logging.error(f"LLM API call timed out for '{plot_label}'.", exc_info=True)
+            return "Sorry, the request to the AI model timed out. Please try again."
+        except aiohttp.ClientError as e:
+            logging.error(f"Error calling LLM API (aiohttp) for '{plot_label}': {e}", exc_info=True)
+            return f"Sorry, I couldn't connect to the AI model at the moment. Network Error: {type(e).__name__}."
+        except json.JSONDecodeError as e:
+            logging.error(f"Error decoding LLM API response for '{plot_label}': {e}", exc_info=True)
+            try:
+                raw_text_response = await resp.text()
+                logging.error(f"LLM API Raw Text Response (on JSONDecodeError) for '{plot_label}': {raw_text_response}")
+            except Exception as read_err:
+                logging.error(f"Could not read raw text response: {read_err}")
+            return "Sorry, I received an unreadable response from the AI model."
+        except Exception as e:
+            logging.error(f"Generic error during LLM call for '{plot_label}': {e}", exc_info=True)
+            return f"An unexpected error occurred while trying to get an AI response: {type(e).__name__}."