Most AI music generators are passive tools driven by text prompts, but BrainJam is built to be a live artistic partner. Instead of typing, the human performer uses their actual physiological and cognitive states as expressive control channels.

The system fuses three biological inputs in real-time:
1. EEG (P300): Reads the performer's visual attention to make discrete musical selections.
2. EMG (Muscle tension): Captures embodied movement to control expressive dynamics like volume and filter sweeps.
3. fNIRS (Cortical blood flow): Tracks slow-state cognitive engagement to modulate the AI’s harmonic tension and complexity over time.

What makes this a true "agent" is the bidirectional feedback loop. The AI (powered by PyTorch/MusicGen) doesn't just generate music autonomously; it continuously adapts its musical proposals based on the human's biological feedback. Built with Python and Streamlit, it’s currently an open-source prototype for my cognitive science PhD research application, exploring the bleeding edge of human-AI co-agency. We are shifting AI from a simple "tool" into an embodied, responsive co-performer.

Stack Used

AI & Core Logic: Google Gemini API (PromptDJ) for semantic musical mapping, PyTorch for real-time sequence generation, and scikit-learn (LDA) for EEG P300 classification.

Biosignal Processing: MNE-Python & SciPy (Signal processing), BrainFlow (Hardware abstraction for EEG/EMG), and Custom Modified Beer-Lambert implementations for fNIRS oxygenation tracking.

Frontend/UX: Streamlit (Web Interface), Plotly (Real-time data visualization), and Custom SVG/CSS for the interactive instrument panels.

Backend & Integration: Python 3.10+, NumPy, Pandas, and LSL (Lab Streaming Layer) for high-precision, low-latency cross-device synchronization.

Generative Audio: MusicGen (Transformers) and MIDI-based symbolic synthesis for zero-latency feedback loops.