À la une
Evaluating AI features before you ship them
A demo is not a launch criterion. How to build an evaluation set, choose a baseline, and decide whether a model is good enough to put in front of users.
TOOKLI Data
Publications
Notes d'architecture, enseignements de livraison et conseils de recrutement — issus de projets réellement menés.
13 articles
- Carrière2 min de lecture
What to ask when you are the one being interviewed
Interviews run in both directions. Six questions that reveal how an engineering team actually works, and what the answers tell you.
TOOKLI Talent
- Web1 min de lecture
Performance budgets that survive contact with a deadline
Every team agrees performance matters and then ships a 2MB bundle. A budget only works if it fails the build.
TOOKLI Engineering
- IA2 min de lecture
Evaluating AI features before you ship them
A demo is not a launch criterion. How to build an evaluation set, choose a baseline, and decide whether a model is good enough to put in front of users.
TOOKLI Data
- Recrutement1 min de lecture
Recruter des ingénieurs sans pouvoir les évaluer soi-même
Un processus concret pour les dirigeants qui recrutent hors de leur domaine — par ceux qui mènent l'évaluation technique.
TOOKLI Talent
- Ingénierie1 min de lecture
Two-week increments, and what they actually require
Everyone claims to work in short cycles. Very few teams can deploy at the end of one. Here is what has to be true before an increment means anything.
TOOKLI Engineering
- Ingénierie2 min de lecture
Pourquoi nous rédigeons une architecture avant tout code
Un document d'architecture est l'endroit le moins coûteux pour se disputer. Voici notre format en six sections, et celle qui évite le plus de reprises.
TOOKLI Engineering
- Données1 min de lecture
The real problem with your dashboards is trust, not tooling
Most organisations do not lack dashboards. They lack agreement on what a number means and a way to tell whether today's figure is wrong.
TOOKLI Data
- Études de cas1 min de lecture
A field app that works where the network does not
Two hundred technicians were losing an hour a day to a web app that failed outside coverage. We rebuilt it offline-first, and let the operations team design the conflict rules.
TOOKLI Engineering
- Mobile2 min de lecture
Offline-first is a data model decision, not a caching one
Bolting a cache onto an online-first mobile app produces conflicts you cannot resolve. Offline-first starts with identifiers and merge rules.
TOOKLI Engineering