akashnaren/agent-ui-sft
Updated • 70
Lab artifacts on which UI is cheapest for agents.
Note Synthetic multi-turn tool traces (80/20)
Note Human-written UI preference labels (50/10)
Note Flat lab-synthetic bakeoff scores (63/14)
Note Pairwise preferred vs rejected UI (65/16)
Note Sklearn lab toy router; see metrics.json
Predict the best UI mode for a text request
Note Static browser demo (not Gradio)