§ 05 — Cases

Six projects. Real numbers, anonymised clients.

What follows are projects I have delivered, with the numbers as they came out of them. The clients are anonymised — I have never named them, and I am not going to start now. Everything below can be walked through in detail on a call, under NDA.

6 documented projects 2019 — 2026 Finance · Retail · Education · Logistics
§ 05.1 — The projects

Six systems that replaced a manual step.

Each one started the same way: somebody showed me the task they did every week and wished it would stop. These are the numbers that came out of it.

LLM API cost optimization Finance
80–100×
lower monthly API spend

Monthly API spend dropped from €12,000 to under €150 while maintaining response quality.

Caching Prompt Engineering Cost Optimization
Customer support AI agent Retail
98%
accuracy with human review fallback

Handles 2,500+ tickets a day. 85% are closed without a human touching them, and the rest go to a person with the draft answer already written.

Fine-tuning RAG
Invoice processing pipeline Finance
40 h → 4 min
manual processing time per week

Reduced manual processing from 40 hours weekly to under 4 minutes with 97% extraction accuracy.

Document AI ETL
Real-time AI dashboard Logistics
94%
adoption among the people it was built for

A logistics firm asked its numbers questions in plain language instead of pulling reports. Query latency dropped 60%, and 94% of the staff who had the tool actually used it.

Dashboard Real-time
Marketing automation engine Retail
+45%
repeat purchases after launch

Automated email campaigns increased repeat purchases by 45%.

Email CRM
Ad cost optimization Education
−60%
Google Ads spend

Reduced Google Ads spend by 60% while maintaining the same lead volume.

Google Ads Optimization
Projects documented here 6 the six on this page, from 2019 to 2026
Support agent accuracy 98% measured on the 2,500 tickets a day it handled
Invoice work per week 40 h → 4 min from 40 hours of manual processing to under 4 minutes
Model cost reduction 80–100× on the API-heavy system taken over from another team
Why there are no client names here
These projects are anonymised. The names belong to my clients, not to me, and most of them have no reason to be on a website because they asked me to build something quiet. The numbers are the part that matters, and they are the real ones. If you want the detail behind any case — the setup, the numbers, what broke on the way — ask me for it on the call and we can go through it under NDA.
§ 05.2 — Start

One of these could be your week.

Send me the task you would rather not do again and I will tell you what it would take to automate it, and what it costs.

Tell me what is slow See what I build