
A submitted assignment or codebase proves an output exists, not that the student understands how it works. Traditional written exams rely on generic question banks, while manual one-on-one vivas require too much faculty time to scale. Meanwhile, AI-authorship detectors are unreliable. Educators needed a scalable way to test whether students can actually explain, defend, and reason about their own work.

I engineered the AI intelligence layer for Enigma—an end-to-end oral assessment platform built around real classroom workflows. Teachers create classes with unique join codes and publish assignments. Once a student attaches their work, Enigma analyzes the document and conducts a personalized voice viva that probes their conceptual reasoning in real time.
To keep orchestration clean and scalable, we decoupled the core web application from the AI compute layer. Next.js and Supabase act as a secure Backend-for-Frontend (BFF) handling authentication, classroom access control, and state transitions, while communicating over strict OpenAPI contracts with my independent Python and FastAPI AI service.
Instead of asking static questions, the FastAPI pipeline parses each student's submitted PDF, extracts key technical concepts and implementation decisions, and generates difficulty-calibrated primary questions. During the session, the engine evaluates each spoken answer and dynamically generates follow-up probes to investigate vague explanations or test edge-case scenarios.
Enigma goes beyond keyword matching by evaluating conceptual accuracy, reasoning depth, and alignment with the submitted work. At the end of every viva, the system compiles a structured diagnostic report detailing demonstrated strengths, partial understandings, specific misconceptions, and targeted recommendations for both students and teachers.
Enigma turns every written assignment into an interactive oral defense without adding manual grading overhead for instructors. By combining classroom management with an autonomous, evidence-based AI examiner, the platform gives teachers clear visibility into genuine student comprehension while giving learners actionable feedback.

Relevant Questions
Concept Mastery Diagnostics
Adaptive Voice Questioning
Client-Side LLM Exposure