Projects

Projects that explore how AI systems can become more observable, testable, and useful in practice.

In progress · Bachelor’s thesis

Current

Automated LLM testing suite

Developed at Lexical Computing, this project turns expected LLM behaviour into a repeatable testing workflow. The aim is to make regression checks and evaluation results easier to run, inspect, and act on.

Problem

LLM behaviour is hard to compare consistently as prompts, models, and product requirements change.

Approach

Capture representative scenarios, evaluate outputs against clear expectations, and make results visible to the team.

Repository and detailed project notes will be linked after public clearance is confirmed.

Concept project · Draft

Example

Scenario lab

A lightweight workspace for comparing models and prompts against the same set of realistic user requests. It pairs a quick quantitative view with examples that make the difference between two runs easy to see.

Question

How can a developer tell whether a change made an experience better, not merely different?

Approach

Run controlled scenarios, annotate notable outputs, and keep a reviewable history of the decisions.

This is draft content used to demonstrate the portfolio format.