Eagerly load embedding model on FastAPI startup

[railway-app]

Problem

The Embedder in backend/indexing/embedder.py lazily loads the all-MiniLM-L6-v2 model via a @Property, deferring the download/initialization until the first request. Since this can take 5-15 minutes, the first requests after deploy hit the backend while it's mid-download, resulting in 502 Bad Gateway errors even though the Dockerfile pre-seeds data via python -m backend.cli ingest sample_docs/.

Solution

Added an explicit load() method to Embedder that forces the lazy model property to resolve. In backend/main.py's lifespan startup handler, we now call app.state.orchestrator.embedder.load() right after constructing the Orchestrator, so the model is fully downloaded and initialized before the FastAPI app finishes startup and begins accepting HTTP requests.

Changes

• Modified backend/indexing/embedder.py
• Modified backend/main.py

---

Generated by Railway