score_model.py

from sentence_transformers import SentenceTransformer, util

_model = SentenceTransformer("all-MiniLM-L6-v2")  # 22 M params, semantic embeddings :contentReference[oaicite:8]{index=8}

def score_fit(text: str, goal: str) -> int:
    emb1 = _model.encode(text, convert_to_tensor=True)
    emb2 = _model.encode(goal, convert_to_tensor=True)
    cos  = util.cos_sim(emb1, emb2).item()           # –1…1 
    return max(0, min(100, int((cos + 1) * 50)))     # map to 0–100