PEPPER

// OVERVIEW

PEPPER is a wearable brain-computer interface that reads EEG brain activity (alpha, beta, theta bands) and heart rate, then uses a six-agent AI pipeline to interpret that state and respond in real time: it switches music to match focus, fires haptic and guided-breathing interventions when stress spikes, and generates post-session focus reports. I built the entire AI pipeline (six services built on Gemini 2.5 Flash and orchestrated with Google's Agent Development Kit) and hand-soldered every component onto the perfboard, including the AD620 instrumentation-amplifier EEG front-end. Built with two teammates over a 36-hour hackathon, it won the Hardware Track at HackaBull VII.

// HIGHLIGHTS

• ›Six-agent pipeline: real-time cognitive-state classification, adaptive music, stress intervention (temple haptics + voice coaching + a guided breathing animation), and a Meeting Agent that generates a personalized "What You Missed" report from the exact moments your attention drifted.
• ›Guardian Agent: a deterministic cooldown layer that throttles interventions so they can't fire too often. Without it, a helpful system becomes more disruptive than the stress it's responding to.
• ›Custom analog EEG front-end: AD620 instrumentation amplifier (~1000x gain) with an all-hardware bandpass filter to isolate the EEG band and reject 60 Hz noise, sampled by an ESP32 at 256 Hz and streamed over a WebSocket.
• ›Real-time DSP: a NumPy FFT extracts alpha/beta/theta band powers from a single EEG channel, driving both the live dashboard and the agents' decisions.