About
Getting better is the product.
Nablo is a private AI workspace that keeps getting better at your company’s work. We build it so every improvement is measured, private and on the record.
How we work
- 01
Measure, then keep.
A change stays only if it scores better on tasks it has not seen. We hold our own research to the same rule.
- 02
Private by default.
Nablo runs where the data lives. We design every part for teams that cannot send data out.
- 03
Show the work.
We publish our experiments with the numbers, including what did not work.
Read the research →
Founder

Isaac Kargar
Founder
Isaac has a PhD in AI and Robotics from Aalto University and more than ten years of AI experience. He is building Nablo to bring that experience to the work companies do with their own information.
Recent research
All research →Prompt Optimization on Tasks Nablo Generated Itself
Nablo generated a verified evaluation from a demo customer's database and optimized its workspace agent's instructions locally. The pass rate on unseen tasks rose from 22% to 67% with an eight-line instruction edit.
Small-Model Distillation, Part 7: Training a 4B Model for Customer Support
Training Qwen3.5-4B on successful customer-support conversations improved its best version’s score, but lower training loss did not mean better results.
Small-Model Distillation, Part 6: Training a 9B Model to Fix SQL Queries
A trained Qwen3.5-9B checkpoint matched GPT-5.5 medium on 220 SQL repair tasks. Teacher examples and verified corrections improved the score, with regressions at every stage.
On the maps
How to read the maps on this site.
Every map here is a contour map of a score. Lines join points with the same score, and the dotted valley is where the best scores are. Each step of the path is a new version: blue steps were kept, red crosses were rolled back.
- 1Same score
- 2Best scores
- 3Version kept
- 4Rolled back
- 5∇Toward better
Now taking on our first design partners.
Start a conversation
Start with one piece of work.
Tell us the work and the data it needs. We will show you how Nablo would learn it inside your environment.
isaac@nablo.io