Engineering Challenges

• Designed a multi-context architecture using Service Workers and Offscreen Documents to bypass Manifest V3 execution limits for long-running computation.
• Eliminated unreliable inter-process communication by moving computation directly to the data instead of transferring large binary payloads.
• Optimized the processing pipeline for predictable job completion rather than maximum concurrency to prevent memory exhaustion.

Engineering Challenges

• Decoupled React state from local storage using optimistic UI updates to eliminate UI stutter when editing massive training plans.
• Implemented a confidence-based parsing system with graceful degradation to safely convert unstructured calendar text into structured workouts.
• Developed a composite algorithmic method validated against golden test fixtures to accurately calculate dynamic pace targets.

Engineering Challenges

• Built an AI workflow without storing user notes or requiring user accounts.
• Treated LLM responses as untrusted input by validating structured AI responses before they reached the UI.
• Explored few-shot personalization without databases or model fine-tuning.