Data Science & AI Technician
I build systems that get small businesses out of the spreadsheet: inventory, automated invoice capture, and the reports decisions actually get made on. I understand the problem, build the project, and leave it running.
01 — Production
What runs today in real companies, with users who depend on it staying up. Each entry opens the case: the problem, the design decision, and what changed afterwards.
Replaced the company's Excel and AppSheet setup with a web app handling stock reservations, shortages, traceability by production area, and reports on turnover and capital tied up.
Invoices arrive by email and are filed on their own: fields are extracted from the PDF, checked against tax rules and recorded, with no typing.
Stock lookups over chat across a catalogue of more than 400 SKUs, so orders can be put together without opening the system or cross-checking spreadsheets.
A web app that generates professional press kits for emerging DJs and producers from a short questionnaire.
02 — Lab
Five end-to-end studies on synthetic data, drawn from the production workshop and the wholesale warehouse where I worked. The data is invented on purpose, so the method can be shown without exposing any company.
How much will sell and when to restock, without deciding by intuition.
SARIMAX with Fourier terms, Prophet and a global LightGBM with recursive forecasting, compared under time-series validation.
A typo in an order is discovered once the item has already been made.
A per-product Z-score rule, Isolation Forest and content-based duplicate detection, evaluated on precision and recall.
Losing a B2B customer becomes obvious only once they are gone.
Out-of-time validation, RFM features with trend, logistic regression against Random Forest and LightGBM, and interpretation with SHAP.
Treating every customer the same wastes the commercial effort.
RFM features with log transformation and scaling, K-Means with K chosen by elbow and silhouette, validated with the Adjusted Rand Index and cross-checked against hierarchical clustering.
Knowing whether a promotional sign works, without mistaking noise for effect.
Sample size and statistical power calculated upfront, blocking by day of the week, and the correct unit of analysis to avoid pseudoreplication.
03 — Profile
Current role
February 2026 — present
I built the reporting module the company bases its purchasing on, and the inventory and orders system it uses every day. I designed the data model, defined the stock alerts, and automated the handoff between production areas.
Before
Data and automation consultancy. Design, development, and the firm's own website.
Inventory auditing with handheld terminals across more than 10,000 SKUs. This is where the operational judgement comes from.
Cash handling, reconciliation, suppliers, and coordinating a team of three.
Education
IFTS N°18 · 2025 — 2027
04 — Contact
If something in your operation is being held together with spreadsheets and patience, it can probably be solved better. Write me what is going on and I will tell you how I would approach it.
If what you need is a consultancy with a team behind it, I work through Renovatio.