About me

AI Engineer & Researcher

Hello! I'm Cecilia Nyberg, an AI researcher and engineer with experience in LLMs, agent systems and deep learning.

As a visiting researcher at the Deep Learning Lab at Seoul National University, I have worked across several areas of AI. My work on the quantized model BitNet b1.58 involved building a 1-bit LLM agent system, analyzing its limitations in long-context and multi-step tasks, and using continued pretraining and supervised fine-tuning to improve Korean language capability. I later shifted toward LLM tool-learning, studying why smaller models struggle with multi-step, domain-specific tool use. This led to a paper on prompt-robust tool learning for LLM-based ship inspection workflows using supervised fine-tuning and prompt augmentation. I am also working on modular CNN training strategies for image recognition, evaluated across six datasets and connected to early-exit network architectures.

On the applied side, I work as an AI Engineer and CPO at LLM Core AI, where I built a full-stack skincare assistant with a LangGraph-based agent architecture, memory, retrieval, streaming responses, Shopify integration, and deployment on Google Cloud Run.

I completed my Master's thesis at Ericsson in 2025, applying machine learning and time-series forecasting to optimize energy usage in radio units. A manuscript based on this work is currently under minor revision at IEEE Transactions on Machine Learning in Communications and Networking.