Cadence

Audience

Learning designers and coaches exploring repeated practice for customer and workplace conversations.

The challenge

Help learners choose and practice useful behavior without reducing a conversation to a script they can copy.

The approach

Combine teacher-authored scenarios with a live character conversation, visible goals, short behavior cues, and a result review tied to learner transcript evidence.

What SideQuest Studio handled

Learning experience design · Scenario and objective authoring · Rive avatar integration · Realtime voice and feedback workflow · Evidence and calibration interface

Important decisions

• Guide the next behavior rather than supplying a copyable response.
• Label scores as estimates and display clear linear meters.
• Check quoted evidence against learner transcript text.
• Keep human-reviewed example capture separate from model training.

Testing and iteration

• Verify lesson creation, conversation turns, speech output, scoring, and results as a complete application flow.
• Check HTTPS mobile layout, accessible controls, and long-lived WebSocket conversation delivery.
• Review adversarial role-switch prompts and the limits of literal evidence matching.

Current outcome

A deployed practice prototype with teacher-authored lessons, live character conversation, and reviewable goal-based feedback.

Known limitations

• Scoring is not a validated assessment and still needs human review.
• Transcript verification does not prove that a quotation demonstrates the intended skill.
• The latest coaching-cue refinement needs a fresh complete lesson review.
• Reviewed examples do not update model weights automatically.

Next steps

• Review complete lessons with representative learners and coaches.
• Evaluate cue timing, variation, and the meaning of quoted evidence.