AI engineer and applied mathematician based in Port Harcourt, Nigeria.
I work on voice AI systems in production and do research on the gaps that production exposes — mainly around evaluation methodology for interactive speech agents and robustness to real-world disfluency.
- Temporal sequence modeling (speech, video, time series)
- Multimodal AI (speech, video, language)
- Evaluation methodology for interactive AI systems
- Numerical methods and PDEs
- Financial mathematics
-
Full-Duplex-Bench v1.0 reproduction — reproduced Lin et al. (2025) for Gemini 3.1 Flash Live on Apple Silicon and ran a systematic analysis of ASR backend sensitivity on turn-taking metrics. TOR varies by up to 62× across backends on identical audio. → paper · code
-
Full-Duplex-Bench v1.5 reproduction — evaluated overlap handling across four scenarios. Gemini correctly responds to user interruptions 72% of the time and shows systematic prosodic adaptation after overlap events. → paper · code
- Founding AI Engineer, Datastrut AI
- MSc Financial Engineering (in progress), WorldQuant University
- First Class Honours, Applied Mathematics, FUTA
- 4 IEEE-indexed publications (2018)
📧 ifeanyi@ifeanyidike.com · 🌐 ifeanyidike.com