Changelog

Version History

Development milestones and release notes for Likhi.

Release history

ℹ️
Likhi follows Semantic Versioning (MAJOR.MINOR.PATCH). Development progresses through defined phases from standalone engine prototype to full production-ready Windows IME.
v1.0.0
Current Target
Production Release · In Development
  • NewProduction-ready Windows installer with full TSF support
  • NewAuto-Correct with strict ≥0.85 confidence gate
  • NewPersonal Dictionary with full CRUD via Settings app
  • NewNative Win32 Settings application with sidebar navigation
  • NewEnglish word preservation in candidate bar for mixed-script writing
v0.5.0
Phase 4
Windows TSF Integration & Suggestion UI Preview
  • NewWindows TSF COM DLL (bangla_tsf.dll) — native in-process input method
  • NewDirectWrite floating candidate window with 1–5 key selection
  • NewP0 keyboard passthrough: Ctrl+C/V/X/Z, F1–F12, Numpad pass natively
  • NewWin+Space language switcher registration as "Likhi (লিখি)"
  • NewGrapheme-aware backspace — zero dangling Hasants
  • NewBengali Dari (।) insertion on period key
  • NewContinuous composition buffer without premature word splitting
  • NewIn-process TSF validation: Notepad, Word, Chrome, Edge, VS Code — PASS
v0.2.0
Phase 3.1
Engine with Lexicon, N-Gram Ranking & Phase 3.1 Optimization
  • New52,161-word binary trie lexicon (lexicon.bin, 4.34 MB)
  • NewLog-linear candidate ranking: phonetic + unigram + bigram + personal dict
  • NewContext-aware bigram transition model
  • NewBanglish fuzzy spelling variation matching
  • NewNon-lexical junk suppression (0.00% invalid candidate rate)
  • ImprovedFlat contiguous binary lexicon: startup 236ms → 3.9ms (98.3% faster)
  • ImprovedRAM: 59.96MB → 13.43MB active (77.6% reduction)
  • ImprovedNext-word prediction latency: 1.42µs
  • New98.08% sentence accuracy on 520-sentence gold benchmark (510/520)
  • New100% Unicode integrity test suite (46/46 pass)
  • New23 complex conjunct stress tests: ক্ষ, জ্ঞ, স্বাস্থ্য, ব্রহ্মপুত্র, উজ্জ্বল
v0.1.0
Phase 1
Standalone Engine Prototype
  • NewPure C++20 core phonetic transliteration engine
  • NewMulti-candidate beam search: Banglish → Bengali phonemes
  • NewFull Bengali Unicode orthography: Swarabarna, Byanjonbarna, Kar, Juktakkhor
  • NewAuto-correct gate (strict ON/OFF) — OFF by default
  • NewStandalone CLI test harness and benchmark runner
  • NewCMake build system with Ninja support
Download Latest →

Powered by