Update utils/utils.py

utils/utils.py

@@ -2,6 +2,8 @@ from PyPDF2 import PdfReader
 from agents.agents import get_agent_groq
 import json
 import re
+import time
+from agents import prompts 
 
 
 def parse_resume(path):
@@ -22,26 +24,92 @@ def parse_resumes(resumes_list):
       resumes_text.append(text)
     return resumes_text
 
+def parse_(resumes_list):
+    resumes_text=[]
+    for resume in resumes_list:
+      text=parse_resume(resume)
+      resumes_text.append(text)
+    return resumes_text
+
+
+from typing_extensions import Annotated, TypedDict, Optional
+
+# Define TypedDict for structured output
+class ResumeAnalysis(TypedDict):
+    candidate_name: Annotated[str, ..., "Name of the candidate with the highest score"]
+    overall_match_score: Annotated[int, ..., "sum of scores for skills_keywords_score, experience_score, education_certifications_score, and preferred_qualifications_score (Whole Number)"]
+    skills_keywords_score: Annotated[int, ..., "Score for Skills and Keywords (0-40)"]
+    skills_keywords_explanation: Annotated[str, ..., "Explanation for Skills and Keywords"]
+    experience_score: Annotated[int, ..., "Score for Experience (0-30)"]
+    experience_explanation: Annotated[str, ..., "Explanation for Experience"]
+    education_certifications_score: Annotated[int, ..., "Score for Education & Certifications (0-20)"]
+    education_certifications_explanation: Annotated[str, ..., "Explanation for Education & Certifications"]
+    preferred_qualifications_score: Annotated[int, ..., "Score for Preferred Qualifications (0-10)"]
+    preferred_qualifications_explanation: Annotated[str, ..., "Explanation for Preferred Qualifications"]
+    score_interpretation: Annotated[str, ..., "donot mention any numbers here, just Interpretation in words of the overall_match_score"]
+
+# Use structured output with the LLM
+
+def generate_analysis_new(resume_text, job_listing_text, job_title_text, must_have, prompt_template):
+    # Send the structured prompt to the agent and expect a structured response
+    agent =  get_agent_groq().with_structured_output(ResumeAnalysis)
+  # using structured output LLM
+    response = agent.invoke(
+        prompt_template.format(
+            resume=resume_text,
+            job_listing=job_listing_text,
+            job_title_text=job_title_text,
+            must_have=must_have
+        )
+    )
+    response['overall_match_score']=response['skills_keywords_score']+response['education_certifications_score']+response['experience_score']+response['preferred_qualifications_score']
+    print(response)
+    return response  # response is already structured as per ResumeAnalysis
 
 def generate_analysis(resume_text, job_listing_text,job_title_text, must_have,prompt_template):
     agent = get_agent_groq()
     resp = agent.invoke(prompt_template.format(resume=resume_text, job_listing=job_listing_text,job_title_text=job_title_text,must_have=must_have))
+    #print('response of agent',resp)
     text_res=extract(resp.content)
     #text_res=extract(text_res)
     #chain = prompt | agent
-    print(text_res)
+    #print(text_res)
     #text = resp.content
     return text_res
 
 def generate_sel_analysis(resume_text, job_listing_text,job_title_text, must_have,prompt_template):
+    prompt_templates = prompts.prompt_template_modern
+    generate_individual_analysis(resume_text, job_listing_text,job_title_text, must_have,prompt_templates)
       #chain = prompt | agent
     agent = get_agent_groq()
     response = agent.invoke(prompt_template.format(resume=resume_text, job_listing=job_listing_text,job_title_text=job_title_text,must_have=must_have))
-    print(response.content)
+    #print(response.content)
     text_res=extract_sel(response.content)
-    print(text_res)
+    #print(text_res)
     return text_res
 
+
+# Analyzing each resume individually and handling delays to avoid token limits
+def generate_individual_analysis(resumes, job_listing_text, job_title_text, must_have, prompt_template, delay=10):
+    #agent = get_agent_groq()
+    all_results = []
+
+    for resume_text in resumes:
+        structured_response= generate_analysis_new(resume_text, job_listing_text, job_title_text, must_have, prompt_template)
+        #agent =  get_agent_groq().with_structured_output(ResumeAnalysis)
+       # print(response)
+        if structured_response:
+            all_results.append(structured_response)
+
+        # Adding delay to avoid the 6000 tokens per minute limit
+        time.sleep(delay)
+
+    # Sorting results by match score (or any other criteria you prefer)
+    best_match = max(all_results, key=lambda x: x.get("overall_match_score", 0))
+    print('best_match',best_match)
+    print('all_results',all_results)
+    return all_results
+
 def extract(content):
     
     json_pattern = r'```\n(.*?)\n```'