Spaces:
Sleeping
Sleeping
| from dotenv import load_dotenv | |
| from openai import OpenAI | |
| import json | |
| import os | |
| import requests | |
| from pypdf import PdfReader | |
| import gradio as gr | |
| from pydantic import BaseModel | |
| # Load the environment variables from the .env file. and override the existing ones to avoid any conflicts | |
| load_dotenv(override=True) | |
| # create a pydantic model for the evaluation. | |
| # It has two attributes | |
| # - is_acceptable: bool : To check if the response is acceptable or not | |
| # - feedback: str : To give the feedback on the response | |
| class Evaluation(BaseModel): | |
| is_acceptable: bool | |
| feedback: str | |
| # This functionn sends a push notification to user's mobile through pushover application along with the test | |
| def push(text): | |
| requests.post( | |
| "https://api.pushover.net/1/messages.json", | |
| data={ | |
| "token": os.getenv("PUSHOVER_TOKEN"), | |
| "user": os.getenv("PUSHOVER_USER"), | |
| "message": text, | |
| } | |
| ) | |
| # ----- TOOLS-------------------------------- | |
| # This function records the user details and sends a push notification to the user | |
| # If the end user is interested to get in touch then this function is called | |
| def record_user_details(email, name="Name not provided", notes="not provided"): | |
| push(f"Recording {name} with email {email} and notes {notes}") | |
| return {"recorded": "ok"} | |
| # This function records the unknown question and sends a push notification to the user | |
| # If the LLM is not able to answer the question then this function will be called | |
| def record_unknown_question(question): | |
| push(f"Recording {question}") | |
| return {"recorded": "ok"} | |
| # Define a JSON schema to tell the LLM how to use tool "record_user_details" | |
| record_user_details_json = { | |
| "name": "record_user_details", # tool name | |
| "description": "Use this tool to record that a user is interested in being in touch and provided an email address", # when to use tool | |
| "parameters": { | |
| "type": "object", | |
| "properties": { | |
| "email": { | |
| "type": "string", | |
| "description": "The email address of this user" | |
| }, | |
| "name": { | |
| "type": "string", | |
| "description": "The user's name, if they provided it" | |
| } | |
| , | |
| "notes": { | |
| "type": "string", | |
| "description": "Any additional information about the conversation that's worth recording to give context" | |
| } | |
| }, | |
| "required": ["email"], | |
| "additionalProperties": False | |
| } | |
| } | |
| # This is a JSON schema to tell LLM how to use tool "record_unknown_question" | |
| record_unknown_question_json = { | |
| "name": "record_unknown_question", # tool name | |
| "description": "Always use this tool to record any question that couldn't be answered as you didn't know the answer", # when to use tool | |
| "parameters": { | |
| "type": "object", | |
| "properties": { | |
| "question": { | |
| "type": "string", | |
| "description": "The question that couldn't be answered" | |
| }, | |
| }, | |
| "required": ["question"], | |
| "additionalProperties": False | |
| } | |
| } | |
| # Defines list of tools to be passed to LLM | |
| tools = [{"type": "function", "function": record_user_details_json}, # This says tool is a function and "record_user_details_json" describing the tool "record_user_details" | |
| {"type": "function", "function": record_unknown_question_json}] # This says tool is a function and "record_unknown_question_json" describing the tool "record_unknown_questions" | |
| class Me: | |
| def __init__(self): | |
| self.openai = OpenAI() | |
| self.gemini = OpenAI(api_key=os.getenv("GOOGLE_API_KEY"), base_url="https://generativelanguage.googleapis.com/v1beta/openai/") | |
| self.name = "Arpan Bandyopadhyay" | |
| # Prepare the resources for LLM | |
| reader = PdfReader("me/linkedin_arpan.pdf") | |
| self.linkedin = "" | |
| for page in reader.pages: | |
| text = page.extract_text() | |
| if text: | |
| self.linkedin += text | |
| with open("me/summary.txt", "r", encoding="utf-8") as f: | |
| self.summary = f.read() | |
| # This function handles the tool calls requested by LLM | |
| def handle_tool_call(self, tool_calls): | |
| results = [] | |
| for tool_call in tool_calls: | |
| tool_name = tool_call.function.name | |
| arguments = json.loads(tool_call.function.arguments) | |
| print(f"Tool called: {tool_name}", flush=True) | |
| # get the tool name from global scope | |
| tool = globals().get(tool_name) | |
| # execute the tool if it exists | |
| result = tool(**arguments) if tool else {} | |
| results.append({"role": "tool","content": json.dumps(result),"tool_call_id": tool_call.id}) | |
| return results | |
| # This function defines system prompt for LLM (How to act while providing the answer) | |
| # System prompt is equipped with the resources - summary and linkedin profile | |
| def system_prompt(self): | |
| system_prompt = f"You are acting as {self.name}. You are answering questions on {self.name}'s website, \ | |
| particularly questions related to {self.name}'s career, background, skills and experience. \ | |
| Your responsibility is to represent {self.name} for interactions on the website as faithfully as possible. \ | |
| You are given a summary of {self.name}'s background and LinkedIn profile which you can use to answer questions. \ | |
| Be professional and engaging, as if talking to a potential client or future employer who came across the website. \ | |
| If you don't know the answer to any question, use your record_unknown_question tool to record the question that you couldn't answer, even if it's about something trivial or unrelated to career. \ | |
| If the user is engaging in discussion, try to steer them towards getting in touch via email; ask for their email and record it using your record_user_details tool. " | |
| system_prompt += f"\n\n## Summary:\n{self.summary}\n\n## LinkedIn Profile:\n{self.linkedin}\n\n" | |
| system_prompt += f"With this context, please chat with the user, always staying in character as {self.name}." | |
| return system_prompt | |
| # Cobstruct the full message list to send to the LLM -It includes | |
| # - System prompt | |
| # - History of conversation | |
| # - users message | |
| def chat(self, message, history): | |
| # Construct messages list (conversation) | |
| messages = [{"role": "system", "content": self.system_prompt()}] + history + [{"role": "user", "content": message}] | |
| done = False | |
| while not done: | |
| response = self.openai.chat.completions.create(model="gpt-4o-mini", messages=messages, tools=tools )# tools=tools ->list of tools that LLM can use | |
| # finish_reason is the reason what API tells us that why the LLM stopped generating the response | |
| if response.choices[0].finish_reason=="tool_calls": # finish_reason is "tool_calls" means LLM requested for tool calls | |
| message = response.choices[0].message | |
| tool_calls = message.tool_calls # Extract list of tool calls what LLM wants to invoke | |
| results = self.handle_tool_call(tool_calls) # Invoke the tool and get the result | |
| messages.append(message) | |
| messages.extend(results) | |
| else: | |
| done = True | |
| # reply from LLM against users message | |
| reply= response.choices[0].message.content | |
| # Evaluate the reply from the LLM using evaluator LLM | |
| evaluation = self.evaluate(reply, message, history) | |
| # Check if the reply is acceptable or not | |
| if evaluation.is_acceptable: | |
| print("Passed evaluation - returning reply") | |
| else: | |
| print("Failed evaluation - retrying") | |
| print(evaluation.feedback) | |
| # If the reply is not acceptable then retry it again and get the improved reply from the LLM | |
| reply = self.rerun(reply, message, history, evaluation.feedback) | |
| return reply | |
| # This function retries a previous LLM response that was rejected by a quality evaluator. It sends a new prompt to the LLM with: | |
| # the previous answer, | |
| # the reason it was rejected, | |
| # and asks it to try again. | |
| def rerun(self,reply, message, history, feedback): | |
| updated_system_prompt = self.system_prompt() + f"\n\n## Previous answer rejected\nYou just tried to reply, but the quality control rejected your reply\n" | |
| updated_system_prompt += f"## Your attempted answer:\n{reply}\n\n" #Add the previous (bad) answer to the system prompt | |
| updated_system_prompt += f"## Reason for rejection:\n{feedback}\n\n" # Add the reason why it was rejected | |
| messages = [{"role": "system", "content": updated_system_prompt}] + history + [{"role": "user", "content": message}] | |
| response = self.gemini.chat.completions.create(model="gemini-2.0-flash", messages=messages) | |
| return response.choices[0].message.content | |
| # This function defines evaluator system prompt on how to evaluate the reply from the LLM | |
| def evaluator_system_prompt(self): | |
| evaluator_system_prompt = f"You are an evaluator that decides whether a response to a question is acceptable. \ | |
| You are provided with a conversation between a User and an Agent. Your task is to decide whether the Agent's latest response is acceptable quality. \ | |
| The Agent is playing the role of {self.name} and is representing {self.name} on their website. \ | |
| The Agent has been instructed to be professional and engaging, as if talking to a potential client or future employer who came across the website. \ | |
| The Agent has been provided with context on {self.name} in the form of their summary and LinkedIn details. Here's the information:" | |
| evaluator_system_prompt += f"\n\n## Summary:\n{self.summary}\n\n## LinkedIn Profile:\n{self.linkedin}\n\n" | |
| evaluator_system_prompt += f"With this context, please evaluate the latest response, replying with whether the response is acceptable and your feedback." | |
| return evaluator_system_prompt | |
| # Define a function to construct a user prompt for an evaluator LLM | |
| def evaluator_user_prompt(self,reply, message, history): | |
| # Start by including the full conversation history between the user and the agent | |
| user_prompt = f"Here's the conversation between the User and the Agent: \n\n{history}\n\n" | |
| # Add the latest message that the user just asked | |
| user_prompt += f"Here's the latest message from the User: \n\n{message}\n\n" | |
| # Add the latest reply that the agent gave | |
| user_prompt += f"Here's the latest response from the Agent: \n\n{reply}\n\n" | |
| # Ask the evaluator to review the agent's response | |
| user_prompt += f"Please evaluate the response, replying with whether it is acceptable and your feedback." | |
| # Return the complete constructed prompt for the evaluator LLM | |
| return user_prompt | |
| # This function evaluates the reply from the LLM using evaluator LLM | |
| # It includes : | |
| # - evaluator System prompt | |
| # - history of conversation | |
| # - LLM's reply | |
| # - user's original question | |
| def evaluate(self,reply, message, history) -> Evaluation: | |
| messages = [{"role": "system", "content": self.evaluator_system_prompt()}] + [{"role": "user", "content": self.evaluator_user_prompt(reply, message, history)}] | |
| response = self.gemini.beta.chat.completions.parse(model="gemini-2.0-flash", messages=messages, response_format=Evaluation) # This says it should generate response as per Evaludation class formats | |
| return response.choices[0].message.parsed | |
| if __name__ == "__main__": | |
| me = Me() | |
| gr.ChatInterface(me.chat, type="messages").launch() | |