Karim Ben BoubakerAI Engineer
I build AI agents people can trust with their knowledge.
I’m exploring how humans and AI agents can collaborate, especially in knowledge work. I build reliable and transparent knowledge systems that help people find, understand, and use domain information with less friction. My work brings together LLMs, symbolic AI, information retrieval, and knowledge graphs.
Selected work
All projects
Sep 2025 — Dec 2025GPT-2 from ScratchBuilt and trained a 35M-parameter GPT-2 model in PyTorch from scratch, with a custom tokenizer, pretraining pipeline, and instruction fine-tuning. Documented the full process from dataset preparation to model training.
May 2025 — Oct 2025Ckeeper: Agentic DevOps PlatformCo-founded an AI-native DevOps startup and helped build an MVP for cloud-aware incident diagnosis, workflow automation, and root-cause analysis. Contributed to product ideation, development, and pitching; the team placed among the top five of 30 teams in the OSTX Bootcamp Ideation Program.
Mar 2025 — Jun 2025Personalized oncology education Q&A systemBuilt a personalized Q&A system for cancer-therapy education with the Tunisian Oncology Association. The system was designed to provide grounded educational answers outside clinical consultations and support oncology teams.
One layer of abstraction at a time
I began my engineering journey in robotics and embedded systems, where I learned how software is built in layers of abstraction: each layer hides the complexity below it so we can focus on what matters. AI adds another layer to that stack, helping us handle more complexity and stay focused on the important questions. I began my career working on this layer in legal technology and regulatory compliance, and more recently in healthcare.
Full experienceWorking on trustworthy AI for knowledge work?
I’m happy to talk about agents, knowledge graphs and evidence-grounded systems.