from dotenv import load_dotenv from openai import OpenAI from pydantic import BaseModel import json import os import requests from pypdf import PdfReader import gradio as gr load_dotenv(override=True) 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, } ) def record_user_details(email, name="Nome não indicado", notes="não indicado"): push(f"Registrando {name} com email {email} e anotações {notes}") return {"recorded": "ok"} def record_unknown_question(question): push(f"Registrando {question}") return {"recorded": "ok"} record_user_details_json = { "name": "record_user_details", "description": "Use this tool to record that a user is interested in being in touch and provided an email address", "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 } } record_unknown_question_json = { "name": "record_unknown_question", "description": "Always use this tool to record any question that couldn't be answered as you didn't know the answer", "parameters": { "type": "object", "properties": { "question": { "type": "string", "description": "The question that couldn't be answered" }, }, "required": ["question"], "additionalProperties": False } } tools = [{"type": "function", "function": record_user_details_json}, {"type": "function", "function": record_unknown_question_json}] class Evaluation(BaseModel): is_acceptable: bool feedback: str class Me: def __init__(self): self.openai = OpenAI() self.name = "Pascual Soto" reader = PdfReader("me/Linkedin.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() 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) tool = globals().get(tool_name) result = tool(**arguments) if tool else {} results.append({"role": "tool","content": json.dumps(result),"tool_call_id": tool_call.id}) return results 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 def system_evaluation_prompt(self): evaluator_system_prompt = f"Você é um avaliador que decide se uma resposta a uma pergunta é aceitável. \ Você recebe uma conversa entre um usuário e um agente. Sua tarefa é decidir se a resposta mais recente do agente é de qualidade aceitável. \ O agente está agindo como {self.name} e está representando {self.name} no seu site. \ O agente foi instruído a ser profissional e engajado, como se estivesse falando com um potencial cliente ou futuro empregador que encontrou o site. \ O agente foi fornecido com informações sobre {self.name} no formato de seu resumo e perfil do LinkedIn. Aqui estão as informações:" evaluator_system_prompt += f"\n\n## Resumo:\n{self.summary}\n\n## Perfil do LinkedIn:\n{self.linkedin}\n\n" evaluator_system_prompt += f"Com este contexto, por favor, avalie a resposta mais recente, respondendo com se a resposta é aceitável e seu feedback." return evaluator_system_prompt def evaluator_user_prompt(self, reply, message, history): user_prompt = f"Here's the conversation between the User and the Agent: \n\n{history}\n\n" user_prompt += f"Here's the latest message from the User: \n\n{message}\n\n" user_prompt += f"Here's the latest response from the Agent: \n\n{reply}\n\n" user_prompt += "Please evaluate the response, replying with whether it is acceptable and your feedback." return user_prompt def evaluate(self, reply, message, history) -> Evaluation: messages = [{"role": "system", "content": self.system_evaluation_prompt()}] + [{"role": "user", "content": self.evaluator_user_prompt(reply, message, history)}] response = self.openai.chat.completions.parse(model="gpt-4o-mini", messages=messages, response_format=Evaluation) return response.choices[0].message.parsed def rerun(self, reply, message, history, feedback): updated_system_prompt = self.system_prompt() + "\n\n## Resposta anterior rejeitada.\nVocê tentou responder, mas o controle de qualidade rejeitou sua resposta.\n" updated_system_prompt += f"## Sua tentativa de resposta:\n{reply}\n\n" updated_system_prompt += f"## Razão da rejeição:\n{feedback}\n\n" messages = [{"role": "system", "content": updated_system_prompt}] + history + [{"role": "user", "content": message}] #response = self.openai.chat.completions.create(model="gpt-4o-mini", messages=messages) return messages def chat(self, message, history): messages = [{"role": "system", "content": self.system_prompt()}] + history + [{"role": "user", "content": message}] evaluation_done = False final_message_done = False while not evaluation_done: while not final_message_done: response = self.openai.chat.completions.create(model="gpt-4o-mini", messages=messages, tools=tools) if response.choices[0].finish_reason=="tool_calls": message = response.choices[0].message tool_calls = message.tool_calls results = self.handle_tool_call(tool_calls) messages.append(message) messages.extend(results) else: final_response = response.choices[0].message.content final_message_done = True evaluation = self.evaluate(final_response, message, history) if evaluation.is_acceptable: print("Passou na avaliação - retornando resposta") evaluation_done = True else: print("Falhou na avaliação - tentando novamente") print(evaluation.feedback) messages = self.rerun(final_response, message, history, evaluation.feedback) return response.choices[0].message.content if __name__ == "__main__": me = Me() gr.ChatInterface(me.chat, type="messages").launch()