import gradio as gr import requests import os # Load API keys securely from environment variables proxycurl_api_key = os.getenv("PROXYCURL_API_KEY") # Proxycurl API key groq_api_key = os.getenv("GROQ_CLOUD_API_KEY") # Groq Cloud API key firecrawl_api_key = os.getenv("FIRECRAWL_API_KEY") # Firecrawl API key class AutonomousEmailAgent: def __init__(self, linkedin_url, company_name, role, word_limit, user_name, email, phone, linkedin): self.linkedin_url = linkedin_url self.company_name = company_name self.role = role self.word_limit = word_limit self.user_name = user_name self.email = email self.phone = phone self.linkedin = linkedin self.bio = None self.skills = [] self.experiences = [] self.company_info = None self.role_description = None # Reason and Act via LLM: Let the LLM control reasoning and actions dynamically def autonomous_reasoning(self): print("Autonomous Reasoning: Letting the LLM fully reason and act on available data...") # Modify LLM reasoning prompt to ask for clear, structured instructions reasoning_prompt = f""" You are an autonomous agent responsible for generating a job application email. Here’s the current data: - LinkedIn profile: {self.linkedin_url} - Company Name: {self.company_name} - Role: {self.role} - Candidate's Bio: {self.bio} - Candidate's Skills: {', '.join(self.skills)} - Candidate's Experiences: {', '.join([exp['title'] for exp in self.experiences])} - Company Information: {self.company_info} - Role Description: {self.role_description} Based on this data, decide if it is sufficient to generate the email. If some information is missing or insufficient, respond with: 1. "scrape" to fetch more data from the company website. 2. "generate_email" to proceed with the email generation. 3. "fallback" to use default values. After generating the email, reflect on whether the content aligns with the role and company and whether any improvements are needed. Respond clearly with one of the above options. """ # Send the reasoning prompt to the LLM url = "https://api.groq.com/openai/v1/chat/completions" headers = { "Authorization": f"Bearer {groq_api_key}", "Content-Type": "application/json", } data = { "messages": [{"role": "user", "content": reasoning_prompt}], "model": "llama3-8b-8192" } response = requests.post(url, headers=headers, json=data) if response.status_code == 200: reasoning_output = response.json()["choices"][0]["message"]["content"].strip() print("LLM Reasoning Output:", reasoning_output) # Now the LLM takes action based on the reflection return self.act_on_llm_instructions(reasoning_output) else: print(f"Error: {response.status_code}, {response.text}") return "Error: Unable to complete reasoning." # Function to act on the LLM's structured instructions def act_on_llm_instructions(self, reasoning_output): # Convert the output to lowercase and trim whitespace for easier parsing instruction = reasoning_output.lower().strip() if "scrape" in instruction: # Action: Scrape the company's website for more information self.fetch_company_info_with_firecrawl() # Reflect again by invoking the LLM to reassess return self.autonomous_reasoning() elif "generate_email" in instruction: # Action: Proceed to generate the email return self.generate_email() elif "fallback" in instruction: # Action: Use fallback logic or default values print("Action: Using fallback values for missing data.") if not self.company_info: self.company_info = "A leading company in its field." if not self.role_description: self.role_description = f"The role of {self.role} involves leadership and team management." return self.generate_email() else: # If the LLM returns an unrecognized instruction, fall back to using the current available data print("Error: Unrecognized instruction from LLM. Proceeding with available data.") return self.generate_email() # Action: Fetch LinkedIn data via Proxycurl def fetch_linkedin_data(self): if not self.linkedin_url: print("Action: No LinkedIn URL provided, using default bio.") self.bio = "A professional with diverse experience." self.skills = ["Adaptable", "Hardworking"] self.experiences = ["Worked across various industries"] else: print("Action: Fetching LinkedIn data via Proxycurl.") headers = {"Authorization": f"Bearer {proxycurl_api_key}"} url = f"https://nubela.co/proxycurl/api/v2/linkedin?url={self.linkedin_url}" response = requests.get(url, headers=headers) if response.status_code == 200: data = response.json() self.bio = data.get("summary", "No bio available") self.skills = data.get("skills", []) self.experiences = data.get("experiences", []) else: print("Error: Unable to fetch LinkedIn profile. Using default bio.") self.bio = "A professional with diverse experience." self.skills = ["Adaptable", "Hardworking"] self.experiences = ["Worked across various industries"] # Action: Fetch company information via Firecrawl API def fetch_company_info_with_firecrawl(self): if not self.company_name: print("Action: No company name provided, using default company info.") self.company_info = "A leading company in its field." else: print(f"Action: Fetching company info for {self.company_name} using Firecrawl.") headers = {"Authorization": f"Bearer {firecrawl_api_key}"} firecrawl_url = "https://api.firecrawl.dev/v1/scrape" data = { "url": f"https://{self.company_name}.com", "patterns": ["description", "about", "careers", "company overview"] } response = requests.post(firecrawl_url, json=data, headers=headers) if response.status_code == 200: firecrawl_data = response.json() self.company_info = firecrawl_data.get("description", "No detailed company info available.") print(f"Company info fetched: {self.company_info}") else: print(f"Error: Unable to fetch company info via Firecrawl. Using default info.") self.company_info = "A leading company in its field." # Final Action: Generate the email using Groq Cloud LLM def generate_email(self): print("Action: Generating the email with the gathered information.") linkedin_text = f"Please find my LinkedIn profile at {self.linkedin}" if self.linkedin else "" # Dynamic LLM prompt prompt = f""" Write a professional email applying for the {self.role} position at {self.company_name}. Use the following information: - The candidate’s LinkedIn bio: {self.bio}. - The candidate’s most relevant skills: {', '.join(self.skills)}. - The candidate’s professional experience: {', '.join([exp['title'] for exp in self.experiences])}. Please research the company's public information. If no company-specific information is available, use general knowledge about the company's industry. Tailor the email dynamically to the role of **{self.role}** at {self.company_name}, aligning the candidate's skills and experiences with the expected responsibilities of the role and the company’s operations. {linkedin_text} Remove references to job posting sources unless provided. Use the LinkedIn URL for the candidate and do not include placeholders. End the email with this signature: Best regards, {self.user_name} Email: {self.email} Phone: {self.phone} LinkedIn: {self.linkedin} The email should not exceed {self.word_limit} words. """ url = "https://api.groq.com/openai/v1/chat/completions" headers = { "Authorization": f"Bearer {groq_api_key}", "Content-Type": "application/json", } data = { "messages": [{"role": "user", "content": prompt}], "model": "llama3-8b-8192" } response = requests.post(url, headers=headers, json=data) if response.status_code == 200: return response.json()["choices"][0]["message"]["content"].strip() else: print(f"Error: {response.status_code}, {response.text}") return "Error generating email. Please check your API key or try again later." # Main loop following ReAct pattern def run(self): self.fetch_linkedin_data() # Fetch LinkedIn data # Let LLM autonomously decide and act return self.autonomous_reasoning() # Define the Gradio interface and the main app logic def gradio_ui(): # Input fields name_input = gr.Textbox(label="Your Name", placeholder="Enter your name") company_input = gr.Textbox(label="Company Name or URL", placeholder="Enter the company name or website URL") role_input = gr.Textbox(label="Role Applying For", placeholder="Enter the role you are applying for") email_input = gr.Textbox(label="Your Email Address", placeholder="Enter your email address") phone_input = gr.Textbox(label="Your Phone Number", placeholder="Enter your phone number") linkedin_input = gr.Textbox(label="Your LinkedIn URL", placeholder="Enter your LinkedIn profile URL") word_limit_slider = gr.Slider(minimum=50, maximum=300, step=10, label="Email Word Limit", value=150) # Output field email_output = gr.Textbox(label="Generated Email", placeholder="Your generated email will appear here", lines=10) # Function to create and run the email agent def create_email(name, company_name, role, email, phone, linkedin_url, word_limit): agent = AutonomousEmailAgent(linkedin_url, company_name, role, word_limit, name, email, phone, linkedin_url) return agent.run() # Gradio interface demo = gr.Interface( fn=create_email, inputs=[name_input, company_input, role_input, email_input, phone_input, linkedin_input, word_limit_slider], outputs=[email_output], title="Email Writing AI Agent with ReAct", description="Generate a professional email for a job application using LinkedIn data, company info, and role description.", allow_flagging="never" ) # Launch the Gradio app demo.launch() # Start the Gradio app when running the script if __name__ == "__main__": gradio_ui()