Proof, Not Promises

Real Problems,Real Systems

Client Work: Legal & Trust Practice

Compliance Screening, Automated

A spreadsheet of names had to be checked against OFAC, ISG, and adverse media sources by hand. It's a strict, repetitive process that took a trained person 6-8 hours per batch. I studied how the analyst actually worked, then rebuilt the same logic as a tool, refined over 17 iterations until it matched the protocol exactly.

6-8h → 1-2h
same result quality, human reviews and signs off every check

Data never leaves the local machine. Same client also had their website built by NeuralSync AI.

Self-Built Project

Nutrition & Training Tracker

A full tracking app: stored pantry and meals, workout logging across multiple sports, water intake, and weight tracking, with calories automatically netted against workout burn instead of juggling several separate apps.

Hours
from idea to a live, working app deployed on Railway

Built to demonstrate build speed as much as build quality.

Client Work: Financial Services

Trustee & Wealth Advisory Website

Full design and build for a Liechtenstein-based trustee firm serving high-net-worth families: practice areas, partner profiles, FAQ, and contact details across three office locations, fully bilingual for an international clientele.

EN / DE
3 office locations, one fully bilingual site

End-to-end build: information architecture, design, and development.

Client Work: Legal Services

Law Firm & Notary Website

Full design and build for a boutique law firm and notary practice operating across Liechtenstein, Austria, and Poland: practice areas, team profiles, FAQ, and multi-jurisdiction contact information, bilingual throughout.

3
jurisdictions served from one site

End-to-end build: information architecture, design, and development.

Self-Built Project

Scanned PDF to Editable Word

Word saves a document as a PDF in one click. Going the other direction, turning a scanned PDF back into an editable Word document, is the harder problem. A local Python tool built around Tesseract OCR (plus an open-source OCR engine and language packs pulled from GitHub) does exactly that, refined over roughly 10 iterations. The build itself took a couple of hours. Getting the output quality to something actually usable took much longer.

90%+
output fidelity, depending on scan DPI

Runs entirely locally. Nothing ever leaves the machine.

Start a Project