The Challenge

CASPer measures empathy, ethics and situational judgement, and there is no answer key. Candidates for medical and professional school were preparing against static materials that could not tell them whether a response was any good, so practice produced repetition rather than improvement. Cracking Med School Admissions needed a platform that could replicate the test environment and then do the harder thing: judge an open-ended written answer the way a human assessor would.

The Engineering Strategy

HashOne Global built CASPer Edge end to end on Next.js 15 with a Node.js and Express backend. Google Gemini generates the feedback, with Qdrant vector search supplying the semantic retrieval that keeps responses grounded in scenario context rather than generic.

The engineering problem was load shape. AI evaluation is heavy and slow, live analytics need to be instant, and both happen while hundreds of test sessions run concurrently. We split them: Redis and BullMQ queue the AI work as background jobs, Socket.IO pushes analytics to dashboards in real time, so a student's progress updates immediately while their feedback generates behind it. Stripe handles recurring subscriptions, Mailgun and MailerLite run transactional and lifecycle email, AWS S3 stores media behind presigned URLs.

The Result

• Every response scored across communication, empathy, ethics and problem solving
• Real-time analytics tracking improvement session by session
• Automated subscriptions, billing and email lifecycle, running without manual intervention

A full-cycle preparation product where the client publishes content through their own CMS and the platform runs itself.

Technology

• Next.js 15
• Node.js
• Express
• Google Gemini
• Qdrant Vector DB
• Redis
• BullMQ
• Socket.IO
• Stripe
• AWS S3
• Mailgun
• MailerLite