The AutoApply Core Engine
AutoApply is the flagship project I have spent the last three years building. It is a highly robust, event-driven agent built on a strict Hexagonal (Ports & Adapters) Architecture.
The Pipeline
- Discovery: Pluggable providers search Google, Bing, Indeed, and direct company pages.
- Vetting: Filters candidates based on commute distance (Haversine formula), skill overlap (SpaCy NLP), and title relevance.
- Application: Uses spatial geometry to parse complex ATS forms (Workday, Greenhouse, Lever), uploads resumes, and uses a local, offline LLM (GPT4All) to dynamically answer open-ended "Why do you want to work here?" questions.
Defense in Depth
To ensure the bot isn't flagged by anti-automation software, it features a heavy evasion framework. It spoofs browser fingerprints, modifies WebGL readouts, uses Bezier curves for mouse movements, and utilizes parabolic timing delays to mimic human keystrokes.
It is built with a "Worst-Case First" philosophy: it is designed to run entirely off a USB flash drive on a 2GB RAM library computer with zero admin privileges.