Back to work / Case study

LexiSync.

A localization collaboration platform combining a PySide6 desktop app, FastAPI services, Vue collaboration UI, AI translation, and QA automation.

RoleProduct / Desktop / Backend / Web
Year2026 - Present
PlatformDesktop / Web / API
Status Active
View repository->
25+
File formats
30+
QA rules
54K+
Approx. code lines
21
Test files

Localization work breaks down when parsing, translation, review, and collaboration live in separate tools.

LexiSync is built for developers, translators, and teams that need to manage multilingual software projects without losing format fidelity or review context.

The product combines local desktop power with optional real-time collaboration so individual workflows and distributed team workflows can share the same foundation.

A desktop-first product with a collaboration layer attached where it matters.

  1. A1Format engine

    A parser registry supports PO, XLIFF, Qt TS, Android XML, Apple strings, JSON, YAML, RESX, subtitles, Office documents, and more.

  2. A2AI translation pipeline

    OpenAI-compatible providers, terminology extraction, translation memory, style guide generation, and review flows support assisted translation.

  3. A3Collaboration server

    FastAPI, WebSocket updates, RBAC, audit logs, chat, and Cloudflare Tunnel make local projects shareable without demanding a complex deployment.

The hardest part is keeping automation useful without damaging source formats.

Parser registryover One-off importers

A shared format abstraction keeps new file types maintainable while preserving format-specific behavior.

Human review loopover Blind AI translation

AI can prefill and suggest repairs, but the workflow keeps translators in control of final text.

Local service sharingover Hosted-only collaboration

Cloudflare Tunnel gives teams a public access path while preserving the desktop-first mental model.

LexiSync turns localization from file editing into a managed engineering workflow.

QA engine

Rules detect placeholder mismatches, punctuation drift, whitespace issues, and other translation defects in real time.

Conflict-aware editing

WebSocket collaboration and Myers diff handling reduce the damage from concurrent edits.

RBAC controls

Admin, reviewer, translator, and viewer roles can be scoped by project, file, and language.

Plugin surface

Custom commands, pseudo-translation, MO reverse compilation, obfuscation, and retrieval enhancements can be added as extensions.

The roadmap is about deeper QA, safer automation, and smoother team workflows.

  • - Expand automated QA repair while keeping edits explainable
  • - Improve translation memory and terminology-assisted AI prompts
  • - Polish the web collaboration experience around review and audit flows
Next case study
DeHalu
2026 / AI code hallucination detection prototype