v1.5.0 — TRACE-CAG benchmark, AI service refactor, speaking practice

What's New

TRACE-CAG Benchmark (AI Service)

• Warm hit rate: 100% (was 65–85%) — P50 cached = 17–29ms (~190x speedup)
• KG enriched: 4,280 → 5,173 concepts (+929 Wikipedia entities từ HotpotQA/2Wiki)
• Model upgrade: llama-3.3-70b-versatile xác nhận hoạt động
• New benchmark modules: cache_utils, env_helpers, kg_utils, llm_client, provider_state, benchmark/

Flutter App

• Speaking answer matcher + FSRS-based speaking practice
• Balanced vocabulary catalog with CEFR A1–C2 policy
• DropdownButtonFormField/Switch UI fixes

AI Service Refactor

• Removed deprecated modules: dual_stream, websocket routes, 12 stale services
• Flutter admin (flutter-admin/) removed — replaced by Vite/React admin dashboard

Backend

• PostgreSQL service for integration tests with healthcheck
• Updated routes: ai_audit, learning, proficiency, progress, user_management, rbac

Dependencies

• pytest-asyncio 1.4.0, celery ≥5.6.3
• react 19.2.7, react-router-dom 7.17.0, @types/react 19.2.17, @types/node 25.9.2
• 20+ dependabot bumps merged

Git Hygiene

• .gitignore updated: data/kuzu_db, data/kg_output (144MB), .crawl4ai, .claude/worktrees
• Dev status report: docs/dev_status_report_2026-06-13.md

Full changelog: dev...v1.5.0

v1.4.0

v1.4.0 — avatar shop, lesson exercises editor, bonus hints, topic cha…

V1.0.2

Full Changelog: v1.0.1...v1.0.2

v0.7.1

Release 0.7.1

v0.5.0

Full Changelog: v0.2.0...v0.5.0

v0.3.0

Release v0.3.0 - Major Feature Update with gamification, proficiency …

v0.2.0

Full Changelog: https://github.com/InfinityZero3000/LexiLingo/commits/v0.2.0