Neurotech · Hardware + software
Turning live brain signals into feedback a client can feel, session after session.
A turnkey neurofeedback system that trained facilitators run with clients: a research-grade EEG headset, a real-time signal-processing engine, and the cloud and fleet tooling to keep every unit running.
- Client
- US neurotech company
- Role
- Lead engineer
- Period
- 2020 – present
- Focus
- Real-time EEG
- in production since
- 2020in production since
- layers: DSP, engine, desktop app, cloud
- 4layers: DSP, engine, desktop app, cloud
- units in the managed fleet
- 9units in the managed fleet
The problem
Neurofeedback only works if the feedback arrives while the brain state it rewards is still happening. Every millisecond between the headset and the sound in the client's ears matters.
The systems are operated by facilitators, not engineers. They have to start reliably, update themselves, and be supportable remotely.
What I built
- 01
Signal-processing engine
A Rust DSP core that computes band envelopes and phase-locking between channels in real time, with artifact rejection so blinks and movement don't get rewarded.
- 02
Neurofeedback engine and session app
A Python engine that turns signal measures into progressive training sessions, inside a desktop app facilitators use to run, score and review sessions.
- 03
Cloud and facilitator portal
A cloud API on Azure for client records and session history, plus a portal where facilitators review sessions and support clients remotely.
- 04
Fleet provisioning
Repeatable provisioning and automatic updates, so a new unit goes from bare machine to running sessions without an engineer on site.
The result
- In production since 2020, through multiple hardware generations and a full rewrite of the signal path.
- Now moving to feedback personalised to each client's own brain rhythms, and driving brain-to-ear latency down toward 100 ms.