app.py

from fastapi import FastAPI
from pydantic import BaseModel
from typing import List
from sentence_transformers import SentenceTransformer
from sklearn.metrics.pairwise import cosine_similarity
import numpy as np

# -----------------------
# Load model ONCE (important for performance)
# -----------------------
model = SentenceTransformer("all-MiniLM-L6-v2")

app = FastAPI(
    title="CV Matching API",
    description="Rank CVs against a Job Description using BERT embeddings",
    version="1.0"
)

# -----------------------
# Request schema
# -----------------------
class MatchRequest(BaseModel):
    job_description: str
    cvs: List[str]

# -----------------------
# Response schema
# -----------------------
class CVScore(BaseModel):
    cv_text: str
    relevance_score: float

class MatchResponse(BaseModel):
    results: List[CVScore]

# -----------------------
# Utility: text cleaning (optional but recommended)
# -----------------------
def clean_text(text: str) -> str:
    return text.replace("\n", " ").strip().lower()

# -----------------------
# API endpoint
# -----------------------
@app.post("/match", response_model=MatchResponse)
def match_cvs(request: MatchRequest):
    # Clean input
    jd = clean_text(request.job_description)
    cvs = [clean_text(cv) for cv in request.cvs]

    # Embed job description
    jd_embedding = model.encode([jd])

    # Embed CVs
    cv_embeddings = model.encode(cvs)

    # Compute cosine similarity
    scores = cosine_similarity(jd_embedding, cv_embeddings)[0]

    # Build response
    results = []
    for cv_text, score in zip(request.cvs, scores):
        results.append(
            CVScore(
                cv_text=cv_text,
                relevance_score=float(score)
            )
        )

    # Sort by relevance (descending)
    results.sort(key=lambda x: x.relevance_score, reverse=True)

    return MatchResponse(results=results)