Hyperreal
Applied AI research & software
Pipelines
Pipelines built to-date:
SzPredict NEW
Open-source seizure prediction benchmark for the CHB-MIT EEG dataset. Standardised evaluation protocols, reproducible metrics, and baseline results. Built to advance clinical EEG research.
FinForm AI LIVE
Financial market prediction using proprietary wavelet-based architecture. 1,100+ live paper trades across 16 cryptocurrency pairs. Built from scratch on local hardware.
VideoMirror LIVE
Assistive video tool for speech therapy. Real-time mirrored camera feed with adjustable delay, designed for children practising articulation and mouth positioning. Free and paid versions on Google Play.
AI News Digest NEW
Daily AI news aggregated from 50 sources, summarised by a local language model running on CPU. No cloud. No tracking. Just signal.
Razr Kade LIVE
Daily AI-generated news show. The News Digest pipeline feeds custom video and image LoRA adapters into a generative video model, producing a fully-synthesised episode every morning. Cyberpunk-satirical voice, no human in the production loop after the curate-and-prompt step.
lorazRadio LIVE
Long-form generative minimal techno. A custom adapter trained on a curated reference corpus shapes an open audio model into a recognisable sonic palette. 150-second tracks, generated locally, published continuously.
Earthwormsoft LIVE
Utility apps on the Microsoft Store and Google Play. PDF tools, file editors and more. 1,500+ users and growing.
HyperReal Labs COMING SOON
AI research portfolio. NeuroWave AI (EEG seizure prediction), FinForm AI (financial market prediction), and experimental architectures built and trained from scratch on local hardware.
Research Roadmap
One core architecture — wavelet-based temporal decomposition feeding sequence models — applied to progressively harder problems. Each stage builds directly on the last.
Seizure Prediction NOW
Predict seizures from scalp EEG before they occur. Cross-patient generalisation on the CHB-MIT dataset — the clinically honest benchmark that most published work avoids. Open-source benchmark framework live.
SzPredict benchmark →EEG-to-fMRI Super-Resolution NEXT
Predict fMRI-quality spatial brain maps from cheap, portable EEG recordings. If temporal patterns encode spatial information — and seizure prediction suggests they do — we can infer deep brain activity from surface electrodes. Democratises neuroimaging: a $200 EEG headset doing the work of a $2M MRI scanner.
Brain-Computer Interface FUTURE
Decode intention from neural signals. The same multi-scale temporal decomposition that distinguishes pre-seizure from normal brain states can learn to distinguish between intended actions, words, or thoughts. Non-invasive, consumer-grade EEG. The logical endpoint of understanding what brainwaves encode.