lever candidate evaluation

main.py

import requests
from requests.auth import HTTPBasicAuth
from environs import Env
import openai

# Initialize environs and OpenAI
env = Env()
env.read_env()
openai.api_key = env("OPENAI_API_KEY")

# Constants
LEVER_API_KEY = env("LEVER_API_KEY")
POSTING_ID = '56db654f-aa40-43f8-a55c-86852bacd064'

# Endpoint, Headers, and Auth
URL_OPPORTUNITIES = "https://api.lever.co/v1/opportunities"
URL_POSTINGS = f"https://api.lever.co/v1/postings/{POSTING_ID}"
URL_RESUMES = "https://api.lever.co/v1/opportunities/{}/resumes"
basic_auth = HTTPBasicAuth(LEVER_API_KEY, '')


def fetch_job_description_from_lever():
    response = requests.get(URL_POSTINGS, auth=basic_auth)
    posting_data = response.json()["data"]
    job_description = posting_data["content"]["description"]
    job_requirements = ' '.join([item["text"] for item in posting_data["content"]["lists"]])
    job_closing = posting_data["content"]["closing"]
    return job_description + ' ' + job_requirements + ' ' + job_closing


def fetch_resumes_for_applicant(opportunity_id):
    url = URL_RESUMES.format(opportunity_id)
    response = requests.get(url, auth=basic_auth)
    if response.status_code == 200:
        return response.json()["data"]
    else:
        print(f"Error {response.status_code} fetching resumes for applicant {opportunity_id}")
        return []


def get_resume_data(opportunity_id):
    URL_RESUMES = f"https://api.lever.co/v1/opportunities/{opportunity_id}/resumes"
    response = requests.get(URL_RESUMES, auth=basic_auth)
    if response.status_code == 200:
        resumes = response.json()["data"]
        # For simplicity, we'll consider the latest (first) resume in the list.
        if resumes and 'parsedData' in resumes[0]:
            return resumes[0]['parsedData']
    return None

def evaluate_applicant(applicant, job_description):
    # Fetch resume data for the applicant
    resume_data = get_resume_data(applicant["id"])

    evaluation_data = f"Applicant Profile: {applicant['headline']} "

    if resume_data:
        for position in resume_data["positions"]:
            evaluation_data += f"Position at {position['org']}: {position['title']} {position['summary']} "
        for school in resume_data["schools"]:
            evaluation_data += f"Education at {school['org']}: {school['degree']} in {school['field']} {school['summary']} "

    # gets hangry
    # if 'linkedin.com' in applicant['links'][0]:
    #     evaluation_data += f"LinkedIn: {applicant['links'][0]} "

    # Direct Lookup URL
    lookup_url = f"https://hire.lever.co/candidates/{applicant['contact']}"

    messages = [
        {"role": "system", "content": "You are a helpful assistant."},
        {"role": "user", "content": f"Given the job description: {job_description}, evaluate the following profile: {evaluation_data}. For example, you might return something like: `{{\"classification\": \"Strong Match\", \"justification\": \"The candidate fits the job requirements well.\", \"email_content\": \"We are impressed with your profile and would like to proceed further.\"}}`"}
    ]

    # Debug print before making the OpenAI call
    print(f"\n\nDebug Info: {messages}")

    response = openai.ChatCompletion.create(model="gpt-4-0613", messages=messages)
    response_text = response.choices[0].message['content'].strip()
    print(f"\n\nResponse: {response_text}")

    try:
        structured_response = eval(response_text)
        classification = structured_response["classification"]
        justification = structured_response["justification"]
        email_content = structured_response["email_content"]
    except:
        # Default fallback in case the LLM does not return the expected structure
        classification = "Unknown"
        justification = "The LLM did not provide a valid structured response."
        email_content = ""

    result = {
        "url": lookup_url,
        "classification": classification,
        "justification": justification,
        "email_content": email_content
    }
    print(f"\n\nResult: {result}")
    import pdb;pdb.set_trace()
    return result

def main():
    job_description = fetch_job_description_from_lever()
    params = {"posting_id": POSTING_ID, "stage_id": "applicant-new", "pipeline": "applicant"}
    response = requests.get(URL_OPPORTUNITIES, auth=basic_auth, params=params)
    if response.status_code == 200:
        applicants = response.json()["data"]
        for applicant in applicants:
            evaluation = evaluate_applicant(applicant, job_description)
            print(evaluation)  # You can expand this to save to CSV or any other format
    else:
        print(f"Error {response.status_code}")


if __name__ == "__main__":
    main()

https://gist.github.com/BenLiyanage/079b3ce7df467879388306962fac734c#file-main-py-L71

Wondering what this does. SHould I be posting the JD as a system message? Also should I be reusing the same conversation for all applicants?