Agent / agents.py
Ahmad140's picture
Update agents.py
75736bd verified
Raw History Blame Contribute Delete
8.85 kB
from langchain_groq import ChatGroq
import json
import streamlit as st
import requests
from bs4 import BeautifulSoup
from utils import embedding_model, get_pinecone_index, save_application
# ---------- Groq LLM ----------
llm = ChatGroq(
model="llama-3.3-70b-versatile",
temperature=0.3,
groq_api_key=st.secrets["GROQ_API_KEY"]
)
class ProfileParserAgent:
def run(self, resume_text):
prompt = f"""Extract structured candidate profile. Return JSON:
{{
"full_name": "...",
"skills": ["skill1", "skill2"],
"experience_years": number,
"education": "degree",
"certifications": [...]
}}
Resume: {resume_text}
"""
resp = llm.invoke(prompt)
try:
return json.loads(resp.content)
except:
return {"full_name": "Candidate", "skills": ["Python", "AI"], "experience_years": 2, "education": "BSc CS", "certifications": []}
class JobSearchAgent:
def run(self, job_titles, locations):
"""
Real-time job search:
1. Try SERP API if API key exists
2. Fallback to free job APIs (Arbeitnow, RemoteOK)
3. Final fallback to mock data
"""
jobs = []
serp_key = st.secrets.get("SERPAPI_API_KEY", None)
# Option 1: Use SERP API
if serp_key:
try:
for title in job_titles:
params = {
"api_key": serp_key,
"engine": "google_jobs",
"q": title,
"location": locations[0] if locations else "Remote"
}
response = requests.get("https://serpapi.com/search", params=params)
data = response.json()
if "jobs_results" in data:
for job in data["jobs_results"]:
jobs.append({
"title": job.get("title", title),
"company": job.get("company_name", "Unknown"),
"location": job.get("location", "Remote"),
"description": job.get("description", "No description"),
"url": job.get("link", "#")
})
except Exception as e:
st.warning(f"SERP API failed: {e}. Trying free APIs...")
# Option 2: Free APIs (if no SERP or SERP failed)
if not jobs:
# Arbeitnow API
try:
resp = requests.get("https://arbeitnow.com/api/job-board-api", timeout=10)
if resp.status_code == 200:
data = resp.json()
for job in data.get("data", []):
if any(t.lower() in job.get("title", "").lower() for t in job_titles):
jobs.append({
"title": job.get("title", ""),
"company": job.get("company", ""),
"location": job.get("location", "Remote"),
"description": job.get("description", ""),
"url": job.get("url", "#")
})
except:
pass
# RemoteOK API (if needed)
if len(jobs) < 3:
try:
resp = requests.get("https://remoteok.com/api", timeout=10)
if resp.status_code == 200:
data = resp.json()
for job in data[:20]:
if any(t.lower() in job.get("position", "").lower() for t in job_titles):
jobs.append({
"title": job.get("position", ""),
"company": job.get("company", ""),
"location": "Remote",
"description": job.get("description", ""),
"url": job.get("url", "#")
})
except:
pass
# Option 3: Mock data (final fallback)
if not jobs:
mock_jobs = [
{"title": "AI Engineer", "company": "HuggingFace", "location": "Remote", "description": "Build AI agents using Transformers, LangChain, Groq. Requires Python, LLMs.", "url": "https://huggingface.co/jobs"},
{"title": "Machine Learning Engineer", "company": "Google", "location": "Remote", "description": "LLM fine-tuning, deployment. Python, TensorFlow.", "url": "https://careers.google.com"},
{"title": "Data Scientist", "company": "Amazon", "location": "Remote", "description": "NLP, transformers, AWS.", "url": "https://amazon.jobs"}
]
jobs = mock_jobs
return jobs[:10] # limit to 10 jobs
class EmbeddingGenerator:
def generate(self, text):
return embedding_model.encode(text).tolist()
class VectorDBStorage:
def __init__(self):
self.index = get_pinecone_index()
def store_jobs(self, jobs):
vectors = []
for i, job in enumerate(jobs):
job_text = f"{job['title']} {job['company']} {job['description']}"
vec = embedding_model.encode(job_text).tolist()
vectors.append((f"job_{i}", vec, {"title": job['title'], "company": job['company'], "description": job['description'], "url": job['url']}))
self.index.upsert(vectors=vectors)
def search(self, query_vec, top_k=5):
results = self.index.query(vector=query_vec, top_k=top_k, include_metadata=True)
return results['matches']
class MatchingRankingAgent:
def run(self, candidate_profile, jobs_with_metadata):
profile_text = f"Skills: {', '.join(candidate_profile.get('skills',[]))}. Experience: {candidate_profile.get('experience_years',0)} years."
profile_vec = embedding_model.encode(profile_text).tolist()
vdb = VectorDBStorage()
vdb.store_jobs(jobs_with_metadata)
matches = vdb.search(profile_vec, top_k=len(jobs_with_metadata))
ranked = []
for m in matches:
ranked.append({
"job": m['metadata'],
"score": m['score'] * 100,
"match_percent": round(m['score']*100, 1)
})
ranked.sort(key=lambda x: x['score'], reverse=True)
return ranked
class CompanyResearchAgent:
def run(self, company_name):
prompt = f"Provide a concise summary of company '{company_name}': culture, tech stack, mission. Keep 80 words."
resp = llm.invoke(prompt)
return resp.content
class ResumeTailoringAgent:
def run(self, original_resume_text, job_description, job_title):
prompt = f"""Tailor the following resume to the job: {job_title}.
Job description: {job_description}
Original resume: {original_resume_text}
Output a revised resume (text format) highlighting relevant skills and achievements."""
resp = llm.invoke(prompt)
return resp.content
class CoverLetterAgent:
def run(self, candidate_name, job_title, company_name, job_description, skills):
prompt = f"""Write a professional cover letter for {candidate_name} applying to {job_title} at {company_name}.
Skills: {', '.join(skills)}.
Job description: {job_description}
Letter should be 150 words."""
resp = llm.invoke(prompt)
return resp.content
class ApplicationAgent:
def run(self, job_url, tailored_resume, cover_letter):
# Simulated submission – in production, use browser automation
return {"status": "simulated_submitted", "submission_id": "hf_demo"}
class InterviewPrepAgent:
def run(self, job_title, job_description, company_name):
prompt = f"""Generate 5 likely interview questions and suggested answers for a {job_title} role at {company_name}.
Job description: {job_description}
Format: Q1: ... A1: ..."""
resp = llm.invoke(prompt)
return resp.content
class NotificationAgent:
def run(self, message, channel="ui"):
st.info(f"📢 {message}")
return f"Sent: {message}"
class AnalyticsAgent:
def run(self):
from utils import get_metrics
df = get_metrics()
if df.empty:
return {"total_applications": 0, "avg_match_score": 0, "submitted_count": 0}
total = len(df)
avg_score = df['match_score'].mean()
submitted = len(df[df['status']=='simulated_submitted'])
return {"total_applications": total, "avg_match_score": avg_score, "submitted_count": submitted}