rag/analysis_chain.py

import os
from huggingface_hub import InferenceClient

from langchain_community.vectorstores import FAISS
from langchain_huggingface import HuggingFaceEmbeddings
from langchain_core.prompts import ChatPromptTemplate
from langchain_core.runnables import RunnablePassthrough
from langchain_core.output_parsers import StrOutputParser
from langchain_core.runnables import RunnableLambda


# =====================================================
#   1. INITIALIZATION SECTION (run ONCE)
# =====================================================

HF_API_TOKEN = os.getenv("HUGGINGFACE_API_TOKEN")
if not HF_API_TOKEN:
    raise RuntimeError("Set HUGGINGFACE_API_TOKEN.")

MODEL_NAME = ""  # When in use insert model name as parameter here

# Initialize remote LLM client ONCE
client = InferenceClient(
    model=MODEL_NAME,
    token=HF_API_TOKEN
)

# LLM wrapper
def hf_llm(prompt: str) -> str:
    response = client.chat_completion(
        messages=[{"role": "user", "content": prompt}],
        max_tokens=400,
        temperature=0.3
    )
    return response.choices[0].message["content"]


# -----------------------------
# Vectorstore initialization
# -----------------------------
def init_vectorstore():
    base_dir = os.path.dirname(os.path.dirname(os.path.abspath(__file__)))
    vector_store_path = os.path.join(base_dir, "data", "vectorstores")

    embeddings = HuggingFaceEmbeddings(model_name="sentence-transformers/all-MiniLM-L6-v2")

    vectorstore = FAISS.load_local(
        folder_path=vector_store_path,
        embeddings=embeddings,
        allow_dangerous_deserialization=True,
    )
    return vectorstore

vectorstore = init_vectorstore()
retriever = vectorstore.as_retriever(search_kwargs={"k": 4})
# =====================================================
#   2. BUILD RAG ANALYSIS CHAIN (no re-init)
# =====================================================

def build_analysis_chain(retriever, llm_callable):
    """
    retriever -> FAISS retriever already initialized
    llm_callable -> function that takes string prompt and returns string
    """
    vectorstore = init_vectorstore()
    retriever = vectorstore.as_retriever(search_kwargs={"k": 4})
    prompt = ChatPromptTemplate.from_template(
    """
You are a professional Resume Analyst AI.
Return a professional summary and hold a conversation keeping the following metrics in mind:

{{
  "job_fit_score": 0-100,
  "fit_summary": "<3 sentence summary>",
  "strengths": ["..."],
  "missing_skills": ["..."],
  "recommendations": ["..."]
}}

RETRIEVED RESUME CONTENT:
{context}

JOB DESCRIPTION:
{job_description}

Analyze and return JSON only.
"""
)


    chain = (
        {
            "context": retriever,
            "job_description": RunnablePassthrough(),
        }
        | prompt
        | RunnableLambda(lambda chat_prompt_value: hf_llm(chat_prompt_value.to_string()))
        | StrOutputParser()
    )

    return chain


# =====================================================
#   3. MAIN FUNCTION FOR ANALYSIS
# =====================================================

def analyze_resume_against_job(job_description: str, retriever, llm_callable):
    chain = build_analysis_chain(retriever, llm_callable)
    return chain.invoke(job_description)

__all__ = [
    "retriever",
    "vectorstore",
    "hf_llm",
    "analyze_resume_against_job",
    "build_analysis_chain"
]

# =====================================================
#   4. Example test run
# =====================================================

if __name__ == "__main__":
    job_desc = "What is the user's machine learning experience?"
    result = analyze_resume_against_job(
        job_description=job_desc,
        retriever=retriever,
        llm_callable=hf_llm
    )
    print("=== ANALYSIS ===")
    print(result)