Stochastic Analytics · Brussels
Data, AI & AutomationConsulting.
An independent practice for data, AI and automation: rigorous statistical modelling, LLM pipelines, and tools that remove repetitive work.
Sebastien Luyasu
Founder · Brussels
Stochastic Analytics is my independent practice, based in Brussels. I'm Sebastien Luyasu: econometrician, statistician, and builder.
I hold a Master's in Econometrics with a research focus in Statistics from Solvay Brussels School of Economics & Management. That training still sets the method: models first, data as the judge, uncertainty stated rather than hidden.
I started out in reinsurance pricing, then worked as an actuary quantifying risk and building automation tools: the kind of environment where a wrong number costs real money.
Today that quantitative rigour goes into two things: client work in data, AI and automation, and independent research in AI safety.
The name and the logo? A logistic curve through a scatter of points. The same function is the statistician's workhorse and AI's original neuron: one foot in each world. The job is what happens in between, finding the curve in the noise.
What we work on
Data, AI and automation at the core, alongside an active research practice in AI safety.
modelling
From raw data to defensible answers: statistical modelling, econometrics and machine learning, built in R and Python. Uncertainty quantified, assumptions stated.
OpenInsurance
A brain for a country's insurance market: an open-source framework that turns insurers' public PDFs into a rich, source-cited knowledge base that any AI agent can read.
Self-sufficient
Clone the repo, add an LLM key, run make all. It scrapes, downloads and extracts from scratch: every input committed, every output regenerable.
Transparent
The exact extraction prompt is a file in the repo, not buried in code. You can rerun the identical extraction with your own model.
Grounded
Every product page cites its source PDF, page numbers included. If it is not in the document, it is not on the page.
A first sample of the Belgian market. Only a fraction of the 50+ insurers so far, and growing.
Plug an agent in, and it can:
- 01compare the exclusions of two policies, side by side
- 02flag duplicate cover when combining two contracts (home + family liability, the classic)
- 03map a market: who sells what, in which branch, from which edition
- 04fact-check its own answer against the source document before replying
Two lines and any MCP client is connected. Keyless, read-only, runs on your machine:
git clone https://github.com/sluyasu/OpenInsurance.git claude mcp add insurance-wiki --env INSURANCE_WIKI_REPO=$PWD/OpenInsurance -- uvx openinsurance-wiki-mcp
Information only, never advice.
How I work.
An estimate without an uncertainty band is an opinion.
Models come first; the data gets the final word.
If it is not in the document, it does not go in the answer.
Automate the repetitive. Audit the automated.
Let's build something sharp.
A data science problem, an AI workflow, an automation, or a digital product: tell me about your challenge.
The first 30 minutes are free. Bring your data.