replace sentence-transformers with OpenAI embeddings — fixes slow HF build

Dockerfile

@@ -1,21 +1,14 @@
 FROM python:3.11-slim
 
-# HF Spaces runs on port 7860
 ENV PORT=7860
 ENV PYTHONDONTWRITEBYTECODE=1
 ENV PYTHONUNBUFFERED=1
 
 WORKDIR /app
 
-# Install dependencies first (Docker cache layer)
 COPY requirements.txt .
 RUN pip install --no-cache-dir -r requirements.txt
 
-# Pre-download the embedding model at build time
-# so cold starts are fast (no download on first request)
-RUN python -c "from sentence_transformers import SentenceTransformer; SentenceTransformer('sentence-transformers/all-MiniLM-L6-v2')"
-
-# Copy source
 COPY . .
 
 EXPOSE 7860

agent.py

@@ -22,13 +22,14 @@ import numpy as np
 import faiss
 import requests
 from openai import AsyncOpenAI
-from sentence_transformers import SentenceTransformer
 
 from knowledge import DOCUMENTS
 import session as session_store
 
 logger = logging.getLogger(__name__)
 
+EMBED_MODEL = "text-embedding-3-small"  # fast, cheap, 1536-dim
+
 # ── Config ────────────────────────────────────────────────────────────────────
 OPENAI_MODEL   = "gpt-4o-mini"
 TOP_K_CHUNKS   = 3
@@ -135,23 +136,32 @@ class SulithaAgent:
         self.name   = "Sulitha Nulaksha Bandara"
         self.openai = AsyncOpenAI(api_key=os.environ.get("OPENAI_API_KEY", ""))
 
-        logger.info("Loading sentence-transformer model...")
-        self._embedder = SentenceTransformer("sentence-transformers/all-MiniLM-L6-v2")
-
-        logger.info(f"Embedding {len(DOCUMENTS)} knowledge chunks...")
+        logger.info(f"Embedding {len(DOCUMENTS)} knowledge chunks via OpenAI...")
         self._doc_chunks = DOCUMENTS
-        embeddings = self._embedder.encode(DOCUMENTS, convert_to_numpy=True, show_progress_bar=False)
-        embeddings = embeddings.astype(np.float32)
+        embeddings = self._embed_texts(DOCUMENTS)
 
         dim = embeddings.shape[1]
         self._index = faiss.IndexFlatL2(dim)
         self._index.add(embeddings)
         logger.info(f"FAISS index built: {self._index.ntotal} vectors, dim={dim}")
 
+    # ── Embeddings (OpenAI) ───────────────────────────────────────────────────
+
+    def _embed_texts(self, texts: list[str]) -> np.ndarray:
+        """Embed a list of texts using OpenAI embeddings. Returns float32 array."""
+        # Use a sync client just for the startup index build
+        from openai import OpenAI as SyncOpenAI
+        client = SyncOpenAI(api_key=os.environ.get("OPENAI_API_KEY", ""))
+        response = client.embeddings.create(model=EMBED_MODEL, input=texts)
+        vecs = np.array([d.embedding for d in response.data], dtype=np.float32)
+        return vecs
+
     # ── RAG ───────────────────────────────────────────────────────────────────
 
-    def _retrieve(self, query: str, k: int = TOP_K_CHUNKS) -> str:
-        q_vec = self._embedder.encode([query], convert_to_numpy=True).astype(np.float32)
+    async def _retrieve(self, query: str, k: int = TOP_K_CHUNKS) -> str:
+        """Embed query via OpenAI, search FAISS, return top-k chunks."""
+        response = await self.openai.embeddings.create(model=EMBED_MODEL, input=[query])
+        q_vec = np.array([response.data[0].embedding], dtype=np.float32)
         _, idxs = self._index.search(q_vec, k)
         chunks = [self._doc_chunks[i] for i in idxs[0] if i < len(self._doc_chunks)]
         return "\n\n---\n\n".join(chunks)
@@ -198,7 +208,7 @@ class SulithaAgent:
         if session_store.is_over_limit(session_id):
             return SESSION_LIMIT_REPLY
 
-        rag_context = self._retrieve(message)
+        rag_context = await self._retrieve(message)
         history     = session_store.get_history(session_id)
 
         messages = (

requirements.txt

@@ -3,7 +3,6 @@ fastapi==0.115.0
 uvicorn[standard]==0.30.6
 pydantic==2.8.2
 openai==1.51.0
-sentence-transformers==3.1.1
 faiss-cpu==1.9.0
 numpy==1.26.4
 httpx==0.27.2