Jarvis Team

Traditional AI assistants explain how to complete tasks, but they rarely verify whether users are actually performing those tasks correctly. This creates a gap between AI-generated guidance and real-world execution. Our project is an AI-powered step-by-step verification system designed to bridge that gap. Rather than simply providing instructions, the system breaks a task into structured steps and uses visual observations to evaluate the user's progress against the expected workflow. The system is designed around four key capabilities: - **Step-by-step verification:** Evaluate whether the user has completed the expected action before proceeding. - **Mistake detection:** Identify incorrect actions, skipped steps, and unexpected application states. - **Outdated tutorial detection:** Recognize when the current interface differs from the original instructions. - **Uncertainty awareness:** Return an UNCERTAIN result when there is insufficient evidence instead of confidently guessing. The project explores applications in software installation, troubleshooting, interactive tutorials, and physical-world tasks. Our central idea is simple: **AI should not just tell users what to do. It should help verify that what they did was correct.** This shifts AI assistance from instruction generation toward evidence-based, verifiable task execution.