src/rag/embedder.py

"""
Embedder.

Same role as the StudyMate version. Returns a NumPy array (not a list) so the
FAISS VectorStore can read `embeddings.shape[1]`. Embeddings are L2-normalized
so inner-product search (IndexFlatIP) == cosine similarity.

Default model is the small general-purpose MiniLM (fast, already cached from
StudyMate). For a code project, a code-aware model retrieves better -- swap in
"jinaai/jina-embeddings-v2-base-code" (pass trust_remote_code=True) once the
pipeline works end-to-end.
"""
import numpy as np
from sentence_transformers import SentenceTransformer


class Embedder:

    def __init__(self, model_name="sentence-transformers/all-MiniLM-L6-v2", **kwargs):
        self.model = SentenceTransformer(model_name, **kwargs)

    def create_embeddings(self, texts):
        embeddings = self.model.encode(list(texts), normalize_embeddings=True)
        return np.array(embeddings)