Ignition is a first-of-its-kind debugging tool built to help both students and professionals understand and refine AI systems by making model reasoning transparent and interactive. Inspired by the struggles students face in learning coding concepts from “black box” AI and the need professionals have for trustworthy debugging in production, Ignition allows users to generate and visualize step-by-step execution plans, edit and re-run tasks, and compare outputs from models like GPT-OSS-120B and GPT-OSS-20B side-by-side. Built with Python 3.9+, PySide6 for the GUI, and powered by llama.cpp and Ollama backends, the tool features interactive graphs, editable notes, and heatmaps that highlight why specific tokens were chosen. Along the way, we overcame challenges such as handling large-scale models, ensuring UI responsiveness, and balancing transparency with safety guardrails. We are proud of delivering a tool that is equally effective in classrooms for teaching coding and reasoning as it is in professional workflows for debugging AI pipelines. Looking ahead, we plan to add teaching modules for educators, team collaboration features, expanded model support, and enhanced explainability dashboards to further bridge the gap between learning and professional development.

Built With

Share this project:

Updates