Pathfinder
Skill Gap Diagnostic Tool
The problem Pathfinder solves
Traditional online learning often creates an illusion of competence. Learners follow step-by-step guides, successfully completing projects without actually mastering the underlying principles.
The Dependency Loop: Learners become reliant on guided instructions and freeze when faced with a blank editor.
Invisible Gaps: Advanced courses assume a level of prerequisite knowledge (e.g., "Closure" in JavaScript or "Backpropagation" in ML) that the student might have missed or forgotten.
Linear Roadmaps: Most platforms offer a "one-size-fits-all" path, ignoring that every learner has a unique "Swiss cheese" knowledge base with different holes.
Challenges we ran into
Building a diagnostic engine is significantly harder than a standard chatbot. Here are the "boss battles" I faced during development:
Socratic Prompt Engineering: It was challenging to prevent the AI from "leaking" the answer. I had to iterate on system prompts to ensure the AI remains a persistent interviewer rather than a helpful assistant.
State Management in Async Loops: Handling the diagnostic "tree" logic required careful state management. If a user fails a question, the backend must instantly pivot to a "sub-topic" without losing the context of the original goal.
Latency vs. Intelligence: Originally, using larger models caused 5-10 second delays. Switching to Gemini 1.5 Flash allowed for sub-second responses while maintaining the reasoning capability needed to evaluate complex technical answers.
CORS & Production Security: Integrating a FastAPI backend with a Vite frontend required fine-tuning middleware to ensure secure data flow while allowing the seamless local development experience.
Technologies used