Independent Research
My AI safety research investigates emergent behaviors in large language models, with a focus on alignment failures that appear during fine-tuning and constrained output generation.
Investigating emergent misalignment in open-weight LLMs. This research examines how seemingly benign fine-tuning can produce unexpected behavioral changes in language models.
Read on arXiv
How forcing JSON outputs doubles misalignment rates in open LLMs. This study reveals how structured output constraints can amplify alignment failures in fine-tuned models.
Read the Research114M+
Users Served at Klarna
10+
Years in Data
PB
Scale Data Systems
I'm a senior Data Analytics & Engineering professional at Klarna, where I turn complex business challenges into production-ready solutions. I build scalable pipelines, automate reporting, and ship AI-powered agents that support analysts across the company.
Off the clock I dive deep into AI safety and fine-tuning research: designing open-weight experiments, benchmarking JSON-constrained agents, and sharing the findings in my publications. That blend of enterprise experience and independent research keeps my toolkit sharp, from Python and SQL to custom eval harnesses for generative models.
Building with AI
Production-ready tools and applications that apply AI to real-world problems, from research automation to accessibility solutions.
Daily AI safety paper feed that filters, categorizes, and summarizes new arXiv research using LLM classification.
View Project
Privacy-first CLI that indexes documents locally and answers questions with cited sources using vector search.
View Project
Live text-to-speech reader that transforms URLs or pasted text into natural-sounding audio with Gemini's multimodal API.
View ProjectTransform raw data into meaningful insights to guide strategic business decisions.
Design and implement reliable data pipelines for efficient data processing and management.
Build assistants, automations, and workflows powered by the latest LLM APIs and vector search.
Lead fine-tuning experiments, evaluation rigs, and safety studies for open-weight and frontier models.
Whether you're interested in AI safety research, enterprise data challenges, or building AI-powered solutions—I'd love to connect.
