A minimal, stripped-down encoder that exists primarily for testing infrastructure rather than production use. It produces text embeddings from very short inputs, with a 512-token context window keeping its scope narrow. Expect no frills — this is a randomly initialized model with no meaningful training behind it.