pages/model2.py

import os
import zipfile
import tempfile
import fitz  
import streamlit as st
from langchain_huggingface import HuggingFaceEndpoint, ChatHuggingFace

os.environ["HF_TOKEN"]=os.getenv('HF_Token')
os.environ["HUGGINGFACEHUB_API_KEY"]=os.getenv('HF_Token')


base_llm = HuggingFaceEndpoint(
    repo_id="meta-llama/Llama-3.1-8B-Instruct",
    provider="novita",
    temperature=0.7,
    max_new_tokens=150,
    task="conversational"
)

llm = ChatHuggingFace(
    llm=base_llm,
    repo_id="meta-llama/Llama-3.2-3B-Instruct",
    provider="novita",
    temperature=0.7,
    max_new_tokens=200,
    task="conversational"
)

def extract_text(file_bytes):
    with tempfile.NamedTemporaryFile(delete=False, suffix=".pdf") as tmp:
        tmp.write(file_bytes)
        text = ""
        doc = fitz.open(tmp.name)
        for page in doc:
            text += page.get_text()
        return text


def resume_prompt(text):
    return (
        "Extract the following from this resume:\n"
        "1. Full Name\n2. Education Background\n3. Total Years of Experience\n4. Technical & Soft Skills\n"
        "5. Key Projects and Outcomes\n\nResume Content:\n" + text
    )


def jd_prompt(text):
    return (
        "Extract this information from the job description:\n"
        "1. Job ID\n2. Company Name\n3. Designation\n4. Required Experience\n"
        "5. Key Skills\n6. Education\n7. Location\n\nJD Text:\n" + text
    )


def match_prompt(jd_text, all_resumes, top_n=3):
    return (
        f"You are an expert recruitment AI. Based on the following job description and candidate summaries, rank the top {top_n} suitable profiles.\n"
        "Evaluate based on: Skills match, Experience, Education, and Project Relevance.\n"
        f"\nJob Description:\n{jd_text}\n\nCandidates:\n{all_resumes}\n\n"
        f"List the Top {top_n} candidates like this:\n"
        "1. Candidate Name - Matching Reason\n2. Candidate Name - Matching Reason\n..."
    )

st.set_page_config(page_title="AI Resume Screener", layout="centered")
st.title("🧠 Smart Resume Screener + JD Extractor")
st.markdown("Upload a ZIP of resumes and a Job Description to discover top-fit candidates, powered by AI.")

zip_file = st.file_uploader("📁 Upload ZIP of Resumes (PDFs Only)", type="zip")
jd_file = st.file_uploader("📄 Upload Job Description (PDF/TXT)", type=["pdf", "txt"])
jd_text_manual = st.text_area("✍️ Or Paste JD Text Directly Below")
top_n = st.slider("🎯 Number of Top Candidates", 1, 10, 3)

if st.button("🚀 Find Best Matches"):
    if not zip_file or not (jd_file or jd_text_manual.strip()):
        st.warning("Please upload both a ZIP file and a JD file/text.")
        st.stop()

    jd_text = ""
    if jd_file:
        jd_text = extract_text(jd_file.read()) if jd_file.name.endswith(".pdf") else jd_file.read().decode("utf-8")
    else:
        jd_text = jd_text_manual.strip()

    st.subheader("📌 Extracted JD Information")
    jd_response = llm.invoke(jd_prompt(jd_text))
    st.markdown(jd_response.content)

    resume_summaries = ""
    with tempfile.TemporaryDirectory() as tmpdir:
        with zipfile.ZipFile(zip_file, "r") as archive:
            pdfs = [f for f in archive.namelist() if f.endswith(".pdf")]
            if not pdfs:
                st.error("No PDFs found in the ZIP file.")
                st.stop()
            st.success(f"✅ Found {len(pdfs)} resumes. Extracting details...")

            for file in pdfs:
                with archive.open(file) as pdf:
                    text = extract_text(pdf.read())
                    summary = llm.invoke(resume_prompt(text)).content
                    resume_summaries += f"\n\n[File: {file}]\n{summary}"

    with st.spinner("🔎 Matching candidates to job description..."):
        final_prompt = match_prompt(jd_text, resume_summaries, top_n)
        match_result = llm.invoke(final_prompt)

    st.subheader("✅ Top Matched Candidates")
    st.markdown(match_result.content)