Seyarkai NeuroSync

The next-gen human-computer interface. A hardware-agnostic, edge-optimized middleware SDK that transforms raw neuromuscular and neural signals (sEMG / EEG) into intentional, high-fidelity control commands for spatial computing.

Disclaimer — read first. NeuroSync is a research-grade SDK. The default signal sources are synthetic generators that simulate wearable sensor streams; real-data ingestion is supported through a documented adapter format. NeuroSync is not a medical device, is not certified for clinical, diagnostic, or safety-critical use, and decoding accuracy on real physiological signals depends entirely on your hardware, electrode placement, and per-user calibration.

What it does

raw sEMG/EEG frames ──► artifact suppression ──► Butterworth band-pass + mains notch
   ──► 200 ms sliding windows ──► Hudgins + spectral features ──► softmax intent classifier
   ──► temporal smoothing ──► `command` events  ( fist / point / flex-wrist / … )

• Ultra-low latency — the per-window decode path (window copy → feature extraction → classification → smoothing) runs in well under 5 ms; a CI perf test asserts p95 < 5 ms. See docs/science.md for the exact measurement methodology.
• Zero-cloud reliance — pure TypeScript, zero runtime dependencies, no network calls. Everything executes locally in the browser tab or Node process. (A C++/WASM core with the same API surface is on the research roadmap; the TS implementation is the reference.)
• Hardware-agnostic — feed any device that produces multi-channel sample frames. Ships with deterministic sEMG/EEG simulators so you can build before you have hardware.
• Per-user calibration — collect labelled windows at runtime, retrain in-process in milliseconds, export/import calibration profiles as JSON.

Install

npm install @seyarkai/neurosync

Browser via CDN (after the package is published):

<script type="module">
  import { Pipeline, SemgGenerator, SignalStream } from
    "https://cdn.jsdelivr.net/npm/@seyarkai/neurosync/dist/index.js";
</script>

Quickstart

import { Pipeline, SemgGenerator, SignalStream } from "@seyarkai/neurosync";

const pipeline = new Pipeline();           // sEMG, 1 kHz, 8 channels, shipped model

pipeline.on("command", (cmd) => {
  console.log(cmd.gesture, cmd.confidence); // "fist" 0.97
});

// Simulated wearable (swap for your own device adapter)
const wearer = new SemgGenerator();
new SignalStream(wearer, (frame) => pipeline.push(frame)).start();

wearer.setGesture("fist");                 // the simulated user clenches

Feeding real hardware instead is one method call per sample:

myDevice.onSample((values: number[], timestampMs: number) => {
  pipeline.push({ t: timestampMs, channels: Float32Array.from(values) });
});

Live demo

neurosync.seyarkai.com — raw vs filtered scopes, live class probabilities, a gesture-driven 3D object, latency HUD, and in-browser calibration. All processing stays in your tab.

Run it locally:

npm run demo        # builds the SDK and serves the demo at http://localhost:8788

Documentation

| Doc | Contents | | --- | --- | | docs/usage.md | Installation, pipeline configuration, events, simulators, real-data ingestion, calibration workflow | | docs/api-reference.md | Every public class, function, type, and event | | docs/architecture.md | Pipeline stages, module map, performance design | | docs/science.md | Filter designs, feature definitions, classifier math, latency methodology | | docs/data-format.md | Frame/CSV schemas for real-device integration | | docs/deployment.md | Production deployment: npm publishing, CI releases, CDN, demo hosting on Cloudflare Pages |

Examples

npx tsx examples/node-quickstart.ts      # simulator → commands in the terminal
npx tsx examples/node-realdata-csv.ts    # decode a CSV recording (real-data adapter)
# examples/browser-minimal.html          # 20-line browser integration

Development

npm install
npm test            # vitest: DSP correctness, features, model, e2e decode, p95<5ms perf
npm run typecheck
npm run build       # tsup → dist/ (ESM + CJS + .d.ts)
npm run train       # regenerate the shipped default model (src/model/weights.ts)

License

MIT © Seyarkai — seyarkai.com