Enterprise AI
I built the savas.one blog pipeline with local models
From article to cover image, I run my whole blog pipeline on my own laptop with fully local models. Here is how it was built and what I learned.
Read articleWriting
Field notes on enterprise AI, cloud and productization.
Enterprise AI
The model reads the invoice, accounting approves it
In procurement an in-house OCR model now reads supplier invoices. The fields come pre-filled, accounting checks and saves. It took 8.1 seconds.
Read articleEnterprise AI
ASUS ProArt PX13 with 128 GB: what can I realistically run at home?
128 GB of unified memory makes the PX13 a solid local AI machine. Great with language models, slow but usable for images, not fast enough for video yet.
Read articleEnterprise AI
From Qwen3.8-27B to Flash-Next: why “Flash” does not mean a small model
Qwen3.8-27B is a practical dense model for local use. Flash-Next is not small: it is a 125B preview that seeks efficiency through a different architecture.
Read articleEnterprise AI
Jev does not chat—it evaluates: where it fits and how to connect it
Jev is not a chatbot. It evaluates a supplied state through predefined Choice, Score and Noul questions, returning structured results an application can use.
Read articleEnterprise AI
Why do enterprise AI projects stall at the pilot stage?
The gap between a demo and a product is rarely the model; it is data, ownership and measurement. We learned this in UAT while working on contracts.
Read articleEnterprise AI
RAG is not a product.
RAG is a key building block for enterprise knowledge access. Without a user problem, permissions, source quality and operations, it stays a technical flow.
Read articleAbout
The personal archive of a product manager who came up through technology.
Notes on enterprise AI, cloud and productization drawn from a career spanning user support, systems operations, DevOps and product management since 2006.
Read the about page