REPNOTES AI
STACK: Swift · SwiftUI · Claude API · Llama 3.1 8B · LoRA · Supabase
PROBLEM
You're between sets with 60 seconds on the clock. Every workout app wants you to tap through menus and dropdowns to log what you just did. Logging a lift shouldn't take longer than doing one.
SOLUTION
One text input. Type "bench 225 3x5" and AI handles the rest, parsing exercise, weight, sets, and reps into structured data instantly.
DEEP DIVE
AI-powered iOS workout tracker using natural language to convert plain text into structured exercise logs. Fine-tuned Llama 3.1 8B via LoRA on Nebius Token Factory using production data across 11 users — built the full ML pipeline including dataset construction, deduplication, quality scoring, and synthetic data generation for underrepresented input patterns. Achieved 95-99% cost reduction and eliminated 10+ second response times versus the Claude API baseline.
HIGHLIGHTS
◆Fine-tuned Llama 3.1 8B via LoRA for 95-99% cost reduction
◆Full ML pipeline: dataset construction, dedup, quality scoring
◆Natural language → structured exercise logs in < 1 second
◆11 beta users on TestFlight