Daedalus is a native SwiftUI application I designed and built for macOS, iPad, and iPhone. It combines a full browser, AI problem-solving, automation, subscriptions, analytics, history, and persistent knowledge into one application and specializes in novel mathematical problems.
Unlike conventional homework solvers limited to predefined questions and input fields, Daedalus is a general problem-solving agent designed to complete and automate novel interactive problems across the web. It is the first system I found designed to autonomously identify, solve, interact with, submit, verify, and learn from both standard and novel problem types across different websites.
I trained and integrated a custom Core ML classifier to identify problem types. I also built a WebKit capture system that isolates the problem from the page and renders content outside the visible screen using PDF and Core Graphics.
The answer-entry system required extensive browser automation. Daedalus scans the live DOM, finds text fields, math editors, multiple-choice controls, buttons, graphs, and draggable objects, then enters and verifies answers using specialized JavaScript and expression-parsing systems. The entire input system is built from the ground up.
I also created DoubleTake, an observe–plan–act–verify agent that can inspect novel pages with multiple interactive steps, perform actions, check what changed, recover from failures, and save successful procedures as reusable machine-readable knowledge. It runs continuously until resolution, and distinguishes Daedalus from other solver apps.
The project required work across SwiftUI, WebKit, JavaScript, Core ML, Vision, Core Graphics, Core Data, StoreKit 2, API networking, Swift Charts, and cross-platform architecture. Building it involved solving hundreds of difficult edge cases involving dynamic websites, structured math input, stale controls, incomplete captures, asynchronous dialogs, etc.
