Codexia Live

The solution

Look for signs that a learner may be stuck, offer a gentle nudge, and let the team inspect the coaching history to improve the experience.

What changed

An initial working product demonstrates timely, progressive coaching. Learner impact and response-time benchmarks are not published.

Lessons learned

Good assistance helps people take the next step without taking over the task.

Behind the solution

Most coding tutors either wait for a prompt or reveal too much. The product needed to detect when a learner was stuck and intervene with the smallest useful hint.

SvelteKit UI streams learner events to a Cloud Run service over WebSockets. The orchestration layer scores friction signals, retrieves relevant exercise context, calls tools when needed, and emits coaching responses back to the UI.

Constraints

• Keep perceived latency low enough for a live learning session.
• Handle partial context, silence, and typing pauses without over-triggering.
• Keep the workflow explainable to product and curriculum teams.

Technical decisions

• Used WebSockets for bidirectional session state instead of polling.
• Kept intervention policy outside the UI so product rules could evolve independently.
• Designed prompts around progressive disclosure to reduce answer leakage.

Key features

• Friction-signal detection from typing pauses and silence.
• Progressive hints with escalation paths.
• Session timeline for debugging agent behavior.

AI system implementation

Learner events travel over WebSockets to Cloud Run. Orchestration scores friction signals, retrieves exercise context, and invokes LLM tools before returning a coaching response. Intervention policy lives outside the UI; progressive prompts start with small hints instead of full answers. Session history exposes coaching decisions for review. No published learner-impact or latency benchmark is claimed.