“Engram's technical approach involves training per-team models within partner workspaces that utilize adapter fine-tuning techniques like LoRAs, prefixes, and sparse architectures.”