// Live signal processing, source localisation and LLM pipelines
— engineered as production systems, not research prototypes
Four disciplines, shipped into production systems rather than left as prototypes
The stack behind the work below
Projects we've built and shipped
An AI-powered audiobook production studio that transforms manuscripts into multi-voice audiobooks. Features LLM-powered script parsing, per-character voice assignment, tonality direction, and a five-phase production pipeline — from raw text to finished audio.
A desktop neurofeedback platform that turns a 16-channel EEG stream into a live picture of brain activity. Signal processing runs in a Kotlin backend at the full 250 Hz sample rate — causal IIR filtering and GEDAI artifact removal on a sliding window — then sLORETA source localization projects the signal onto 8,196 cortical sources, rendered as a 3D brain in the browser. Operators get electrode impedance checks, live band power, network connectivity and reward-based training in one workspace.
A review and reputation management platform with NLP-powered sentiment analysis — built years before AI went mainstream. Aggregates reviews from Google, Facebook, and other platforms with real-time analytics and automated collection via mobile.