Case study · AI agents · Live in production
AI cheating detection for technical interviews
Do Trung Nghia led a real-time engine that flags AI-generated answers during live technical interviews, in production at neo.speedforce.ai.
Since AI assistants arrived, the take-home answer and the interview answer stopped being written by the same person. Hiring teams needed a signal inside the interview itself, in real time, without turning the session into an interrogation.
The system
A real-time engine that scores interview responses as the conversation happens. It runs alongside the interviewer, not instead of them: the interviewer sees a confidence signal and keeps the final call. Led by Do Trung Nghia as tech lead, team of 10, in production at neo.speedforce.ai.
Design decisions that mattered
- Confidence scoring over binary verdicts. A single "cheating" flag would decide someone's career on one number. The engine reports calibrated confidence so a human makes the decision.
- Interviewer UX first. The signal had to fit into the interview flow without distracting the interviewer from the candidate.
- Privacy and legal constraints up front. What is monitored, what is stored, and what can be shown to whom were designed as constraints of the system, not afterthoughts.
FAQ
- How does AI cheating detection work?
- The engine analyzes interview responses in real time and reports a confidence signal when answers look AI-generated. The interviewer, not the engine, makes the call.
- Is it live?
- Yes, in production at neo.speedforce.ai.
- Who built it?
- Do Trung Nghia, as tech lead of a team of 10: confidence scoring design, interviewer UX, and the privacy and legal constraints around the system.
- How do we start a similar build?
- Email work@dotrungnghia.com with the problem, your data, and the deadline. See AI consulting services.