The recent advancements using state-space models enable greater context windows (Jamba with 256k context) and capability of learning over time. We propose to use continuous reinforcement learning to enable a fine-tuned Jamba instance to learn from responses of identified senior members to engage helpfully in a group chat, eg our team slack, and the NEAR discord. Our goal is a focused expert AI that helps small teams guide new members in alignment with the team code, structure and mission.