projects

Production AI systems, research projects, and open-source work.

Research

Full-Duplex Conversational AI — Benchmark Reproduction & Extension

Reproduction and extension of Full-Duplex-Bench (Lin et al., 2025), evaluating turn-taking and overlap handling in voice AI systems.

Couette-Poiseuille Reactive Fluid Flow with Variable Thermophysical Properties

Applied mathematics research on modeling reactive fluid flow with variable viscosity and thermal conductivity using coupled nonlinear PDEs, perturbation methods, and finite difference numerics.

ETF Trend Prediction in Emerging Markets — Paper Reproduction

Reproduced Sagaceta-Mejía et al. (2024) using LASSO feature selection and MLP neural networks to predict ETF trends in Chile, Brazil, and U.S. equity markets. WQU MScFE coursework.

Open Source

AI Testimonial Intelligence Platform

Built a multimodal AI platform combining speech-to-text, sentiment analysis, and video understanding to automate testimonial insight extraction and story video composition at Cenphi.

TubeForge — Multimodal AI Generation Pipeline

End-to-end multimodal generation system producing temporally-aligned speech, visual, and language outputs from a single text prompt. Validated at scale — grew a faceless YouTube channel to thousands of subscribers autonomously.