Independent engineer
building local-first
tools and AI infra
Portfolio
2026 / v01
Open to engineering work2026hello@tarxf.dev >

Full-stack&Systems-02

I build local-first tools, desktop software, AI infrastructure, and i18n systems, from product surfaces down to sync, search, and release pipelines.

Scroll
Local time
--:--:--
About / 01
(01) - Introduction

I work where product meets infrastructure.

I build ambitious tools across desktop software, full-stack systems, AI infrastructure, and localization workflows. Most of my work starts with a real workflow problem and turns into a product with a local-first data model, a careful interface, and enough infrastructure to keep growing.

My projects tend to connect product engineering with systems thinking: PySide6 desktop apps, TypeScript front ends, Go or FastAPI services, SQLite search paths, edge APIs, and AI/RAG pipelines that are useful because they are grounded in real constraints.

- Currently in Tianjin, CN
(02) - Process

How I work.

The common thread is simple: understand the workflow, design the data path, build the usable layer, then measure where the system starts to strain.

  1. 01
    Model

    Start by mapping the entities, constraints, and failure modes. A good interface is easier to build when the data model already tells the truth.

    Workflows / Data / Constraints
  2. 02
    Prototype

    Build the smallest version that proves the hard part: sync behavior, search latency, parser correctness, retrieval quality, or integration boundaries.

    Spikes / Benchmarks / Thin slices
  3. 03
    Engineer

    Turn the prototype into something maintainable: clear modules, predictable APIs, test coverage on hot paths, and careful degradation when real platforms get messy.

    Architecture / Tests / Tooling
  4. 04
    Ship

    Package the work as a usable product: installers, documentation, release metadata, localized copy, and the finishing details that make tools feel trustworthy.

    Release / Docs / Feedback
(03) - Selected work

Selected work

(06) / 2024 - 2026
01
HistorySync
A privacy-first browser history platform that unifies desktop search, browser extension access, and multi-device incremental sync.
PySide6SQLite FTS5WXTReactGo Fiber
2026
02
LexiSync
A localization collaboration platform combining a PySide6 desktop app, FastAPI services, Vue collaboration UI, AI translation, and QA automation.
PySide6FastAPIVue 3WebSocketSQLite
2026
03
DeHalu
A prototype-stage AI code hallucination detection platform using MCP gateways, streaming checks, RAG, GraphRAG, and governance pipelines.
HonoTypeScriptPythonPostgreSQLGraphRAG
2026
04
NovaSort
A C++20 header-only adaptive hybrid sorting library tuned for real input distributions, structured data, and benchmark-driven performance work.
C++20CMakeIntrosortTimSortFuzzing
2026
05
l10n-expansion-data
An open localization dataset that estimates text expansion ratios by language pair, source-length bucket, and statistical confidence level.
OPUS-100JSONCSVYAMLStatistics
2026
06
HistorySyncWeb
The product website and download distribution system for HistorySync, combining an Astro marketing site with a Cloudflare Worker release API.
AstroTypeScriptTailwind CSSCloudflare WorkersREST API
2026
Currently

Now.

Updated / June 2026
Building
HistorySync, LexiSync, and the surrounding release infrastructure for local-first productivity and localization workflows.
Exploring
AI coding reliability, hallucination detection, GraphRAG, MCP gateways, and evaluation loops that make agent output safer to trust.
Measuring
Sorting behavior across real input distributions, browser-history search latency, and localization layout risk before it reaches the UI.
Open to
Full-stack engineering, desktop software, AI infrastructure, and i18n tooling.
Let's build
systems that
earn trust.
- GitHub- Read.cv- X / Twitter- LinkedIn