Spark

I am leading EduLLM (Beyond the ChatBot) under the SNSF Spark grant to rethink how large language models can support non-computer science students as they learn introductory programming. The goal is to move past generic chatbots and design evidence-based interactions that respect pedagogy, ethics, and students’ confidence. Why this project Many learners rely on ad-hoc chatbot answers that can be misleading or counter-productive. Educators need guardrails that align with their course design, assessment policies, and academic integrity expectations. HCI-driven experimentation can show where generative feedback helps or harms early-stage learners. What we’re building LLM-mediated learning flows that scaffold problem-solving steps instead of providing full solutions. Ethical and transparency layers so students understand model limits, provenance of hints, and acceptable use. Instructor controls to tune guidance levels, restrict disallowed prompts, and surface analytics about help-seeking behaviors. Benchmarks and rubrics tailored to novice errors in Scala and Python exercises to evaluate LLM responses. Research approach Mixed-method classroom studies combining telemetry, think-aloud protocols, and graded outcomes to measure effectiveness and trust. Iterative prototyping with rapid A/B comparisons of prompt strategies, moderation filters, and UI cues. Funding and collaboration Funded by the Swiss National Science Foundation (Grant No. 228765). If you’re teaching introductory programming and want to collaborate on trials or share datasets of student questions, feel free to reach out.