Skip to content
ZYNXTECH

About

Tools, apps and reports, published with their source.

Zynxtech is the name behind 8 open-source projects: developer tools for AI coding agents, full-stack applications for personal finance and HR, and Power BI reports. Each one is built to be used, read and checked by someone other than its author.

There is no product to sign up for and no hosted service. The output is source code, packages on PyPI and npm, and the documentation that goes with them.

01How the work is done

  1. 01

    Run on the user’s machine

    LLMDyno benchmarks on your own GPU and keeps your data in a local folder. pbidriver drives Power BI Desktop without publishing anything. The applications are self-hosted. Nothing depends on a Zynxtech server.

  2. 02

    Publish the method with the number

    A score comes with its weights, an estimate with its measured error, a result with the transcript that produced it. LLMDyno states that its VRAM projection is off by 66 MiB at worst across 84 configurations, and says when a gap is too small to mean anything.

  3. 03

    Write the limits down

    Each README has a section on what the project does not do: which Power BI versions the selectors were tested on, which visuals cannot be read, which defaults to change before a real deployment.

  4. 04

    Ship the documentation with the code

    Documentation sites, API references generated from the code, setup guides for each coding agent, and an llms.txt so AI assistants can read the docs as well.

  5. 05

    Let tests hold the promises

    When the docs claim two things stay in step, a test checks it: LLMDyno’s library, CLI and web API, pbidriver’s Python and npm version numbers, the README links that must work on PyPI.

  6. 06

    Open by default

    Every project on this site is MIT licensed, with its full source on GitHub.

02The author

Jerome Eliezer

Author of every project here

Jerome Eliezer writes the code, tests and documentation for every Zynxtech project, and maintains the releases on GitHub, PyPI and npm.

The projects show the range. LLMDyno is a Python service with a FastAPI back end, a React and ECharts dashboard, Docker sandboxes for seven coding agents and a memory model for GGUF files that lands within 66 MiB of llama.cpp's own fit check. pbidriver drives Power BI Desktop through the Chrome DevTools Protocol, Win32 messages and the ADOMD.NET client. Finventory carries an accountant’s rules for balances and an Indian tax engine. The Power BI reports come from the other side of the same work: the analyst’s view of the numbers.

The recent work centres on AI coding agents, and on giving them the instruments they need to check their own results.

03Tools in use

Languages
Python, TypeScript, JavaScript, DAX, SQL
Back end
FastAPI, Django, Django REST Framework, Flask, SQLAlchemy, Celery
Front end
React, Vite, Tailwind CSS, TanStack Query, ECharts, Recharts
Data
PostgreSQL, SQLite, Redis, Power BI, Power Query
Automation
Playwright, Chrome DevTools Protocol, WebView2, Win32
Models and agents
llama.cpp, GGUF, OpenAI and Anthropic APIs, agent harnesses
Infrastructure
Docker, GitHub Actions, trusted publishing to PyPI and npm, MkDocs, Cloudflare

04This site

Built with Astro, TypeScript and Tailwind CSS, and served as static files from Cloudflare Pages. Fonts are self-hosted. There are no cookies, no analytics scripts and no third-party requests. The project pages are written from each repository’s README, changelog and source, and the Power BI pages from the reports’ own layout files.