PathLock

Why we built this

Traditional online learning suffers from two critical problems: generic one-size-fits-all curricula and weak verification of actual understanding. Learners are bombarded with expensive courses that don't match their skill level or goals, and completion certificates only prove someone clicked through videos, not that they actively learned. Meanwhile, the rise of AI tools like ChatGPT has made it trivially easy to fake understanding by submitting AI-generated summaries. For career changers and skill-seekers, this creates a frustrating cycle of consuming content without building real competence. We believe that personalized learning paths combined with AI-powered verification of understanding is crucial for making self-directed education meaningful and trustworthy.

What it does

PathLock creates personalized learning paths through conversational AI and verifies true understanding with an AI-powered evaluation framework. Unlike traditional learning platforms that offer static courses with multiple choice questions, PathLock dynamically generates customized learning journeys based on your goals, experience level, and time commitment. The learning journey isn’t AI generated, it’s a list of human sources for the user to actively learn from. Each topic is "locked" until you demonstrate real comprehension by submitting notes that are evaluated against key concepts, with built-in detection for AI-generated or plagiarized content. This gated approach ensures learners truly master each concept before progressing, making the learning path a credential that actually means something.

How we built it

PathLock was built using a modern TypeScript stack with Next.js 16 and React 19 on the frontend, styled with Tailwind CSS 4. The backend leverages Next.js API Routes with Supabase (PostgreSQL) for data persistence. We integrated Anthropic's Claude (claude-sonnet-4-5) for conversational path planning and note evaluation, and Tavily Search API for curating authoritative, free learning resources. Two specialized AI agents power the platform: a Path Planning Agent that conducts conversational needs assessments and generates structured learning paths, and a Gated Learning Agent that researches materials, extracts key concepts, and evaluates student submissions with plagiarism detection. Real-time progress updates are delivered via Server-Sent Events (SSE) streaming. Keywords AI analytics are built into both agents, with different tasks tagged for evaluation.

Challenges we ran into

Building reliable AI agents that could handle complex, multi-step workflows was challenging. We needed to balance between quality and feedback time by carefully designing tool calls and the RAG process. Curating quality learning materials meant filtering out paid courses, listicles, and paywalled content while ensuring resources were genuinely educational. Implementing weighted concept coverage scoring where critical concepts matter more than supplementary ones required careful calibration to create fair but rigorous evaluations.

What we learned

This project deepened our understanding of agentic AI architectures, particularly how to design reliable multi-step workflows with appropriate guardrails. We learned the nuances of prompt engineering for structured output generation versus open-ended conversation. Working with streaming APIs taught us patterns for providing real-time feedback during long-running AI operations. Perhaps most valuable, we gained insight into the challenges of educational assessment—balancing rigor with fairness, detecting dishonesty without false positives, and creating evaluation criteria that reward genuine understanding over rote memorization.

What's next for PathLock

We plan to expand PathLock with more customization features, like curating more to each individual’s learning preferences and history. Advanced analytics could help learners identify knowledge gaps and optimize study time. We want to integrate with existing learning management systems so educators can use PathLock's evaluation engine for their courses. Additionally, we plan to add verified learning portfolios that showcase completed topics with source materials and demonstrated competencies—providing tangible proof of genuine understanding. This positions us to attract learners who are dissatisfied with current AI tools and want credentials that actually reflect real skill acquisition.