WORK

Real systems.
Real numbers.

Five builds, described with the numbers from the published case studies. No client names. The AI receptionist is an open-source reference build, not client work.

01 / Legal tech

From court judgments to structured data.

Long, inconsistent judgments turned into clean fields a legal product runs on.

955: judgments in the first full volume
39: fields extracted per case
23: pipeline steps, run from one config

Python
SQLite
Local LLMs
pytest

02 / Government agency

Six to eight hours of manual work. Now automatic.

Incoming Excel files detected, classified and routed without anyone opening them.

6–8 h: manual job, replaced
Every: exchange logged for audit
Failed: records set aside with a reason

Python
Apache NiFi
MongoDB
PostgreSQL
Keycloak

03 / Music tech

BigQuery to Snowflake, without losing trust.

A migration with an automated check that every table arrived complete and correct.

SQL: checks generated on the fly
Airflow: schedules for every move
Live: dashboards on progress

Python
Snowflake
dbt
BigQuery
Airflow

“

[He] also demonstrated great initiative by designing an automated validation system to ensure data consistency between Google and Snowflake migrations.

VPMusic-tech company

04 / Operations

Reporting that runs itself.

Spreadsheets and forms turned into small systems with structure behind them.

Auto: updating reports
1: form-based inventory
3: messy source formats cleaned

Power BI
Apps Script
TypeScript
Python

05 / Reference build · Agentic AI

An AI receptionist connected to real systems.

Open source, built around a fictional clinic: it checks insurance and books real slots.

12 / 13: caller scenarios passed first run
1.4 s: median reply on a recorded call
$0.05: model cost per two-minute call

Voice AI
Open source