Abdur Rahman Eshan

Draw2Verify is an AI agent that won't give you an unverified circuit answer. The problem: AI can read a photo of a hand-drawn circuit, but it can misread a value or a connection and still answer with full confidence. For students and engineers, a confident wrong answer is worse than no answer. How it works: you upload a photo of a hand-drawn circuit and ask a question by typing or speaking (Muse Voice). Muse Spark works as an agent: it picks one tool at a time to build a checklist, zoom into the image, build an editable circuit graph, ask you when something is unclear, and run ngspice simulations (DC, sweep, time and frequency). Every number is cross-checked by an independent solver. It can also test a proposed fix without changing your original design. Before any answer is marked verified, Muse Wonder independently re-reads the photo without seeing the agent's answer, the two circuits are compared, and every claim is judged against the evidence. Failed checks go back to the agent as feedback for up to 3 correction rounds; a failure is never turned into a pass. Unsupported circuits (e.g. transistors) are honestly reported as unsupported. You can download the verified design as a schematic PDF, a parts list and a reproducible simulation package. Evaluation: on 20 held-out hand-drawn circuits with frozen code, all parts and connections were found correctly (18/18) and numbers matched in 15/17. On 5 extra wider-scope cases, all 4 supported circuits were verified correct and the transistor case was correctly refused. Built solo during the hackathon with Meta's Muse Code. It uses Muse Spark (agent and reader), Muse Wonder (blind second reading and claims judge) and Muse Voice (speech to text).