The stories behind building Vinci.
From Toronto, we publish our decisions, the evidence behind them, and what we learned along the way.
Newest first
- 6 min readGeorge Pu
Why Vinci started with post-training
We want to pre-train our own 4B and 10B models. George shares why that matters for Canadian AI, what the economics look like, and why post-training came first.
Read - 2 min readGeorge Pu
Introducing Vinci MLE 123B 1.0
Our largest MLE model is now available with BF16 and GGUF weights. Built on Devstral 2, it extends Vinci’s work on models for machine learning engineering.
Read - 8 min readGeorge Pu
We tried a new AI architecture. It didn't work.
Our first native architecture experiment fell short on basic language modelling. We stopped it, kept the useful engineering, and changed how the next idea will earn more compute.
Read - 3 min readGeorge Pu
Vinci is focusing on models and verification
What we learned from our first six weeks, and what we are building next.
Read - 6 min readGeorge Pu
Beyond fine-tuning: what Vinci is becoming
Fine-tuning was enough to change our models' voice. It was not enough to build the models we now want to release. We are moving from light behaviour tuning to full-weight model redevelopment - open foundations, rebuilt by Vinci. Here is the plan, stage by stage.
Read
We publish the parts most companies cut.
Real numbers
What inference costs us, what a decision cost us, what we charge and why. If we cite a number, the source is in the post.
Decisions with their reasons
Not just what we chose — the option we rejected, and the constraint that made the call. Including the ones we'd take back.
The things that broke
Deploys that failed, assumptions that didn't survive contact with users, and features we pulled. Written down before they're comfortable.
Want the longer version?
George's personal essays on building this company live on Founder Reality — separate site, longer form, more opinionated. This log stays about the product.