ECE · Delhi Technological University
Abhishek Kumar
Gupta
Building wearable physiological sensors, 1000 Hz real-time DSP, and auditing emotion AI models.
Scroll
Building wearable physiological sensors, 1000 Hz real-time DSP, and auditing emotion AI models.
Every past study and circuit prototype was a deliberate node built to solve one piece of the puzzle. Together, they form the constellation that powers Novara.
Exposed video ID leakage where models hit 83% without biosignals. Proved wearable AI fails without ground-truth physical signal fidelity.
Breadboarded AD8232 & MAX30102 front-ends with custom Python oscilloscope telemetry. Solved dual-trace timing and Pulse Transit Time (PTT).
The apex where all foundational research unites: a continuous clinical-grade wearable engineered to catch hypertension and maternal preeclampsia before clinical crises.
Analyzed 5,959 trials at 1000 Hz. Proved heart slows by -1.93 BPM during mental focus and engineered <15ms Pan-Tompkins edge detection.
Audited 42 multi-day continuous cycles. Modeled subtle biological shifts (d = 0.27), replacing static alarms with personal baseline tracking.
Bridging physical analog bio-potential front-ends with sub-millisecond digital signal processing.
Continuous-time filter design, QRS complex extraction, artifact rejection, and time/frequency HRV metrics extraction.
Ultra-low-power firmware, high-speed SPI/I2C sensor buses, 4-layer PCB layout, and analog ground plane isolation.
Mixed-Effects modeling, feature engineering across autonomic rhythms, and robust cross-subject validation schemes.
In-depth understanding of autonomic nervous system dynamics, cardiac electrophysiology, and cognitive workload states.
Bridging non-invasive physiological bio-telemetry with continuous cognitive state awareness.
Most consumer wearables stop at step counts and rough daily averages. I focus on building low-latency hardware and signal processing pipelines that capture real-time physiological dynamics as they happen—from subtle cardiovascular slowing during deep concentration to early warning indicators of autonomic strain.
Integrating high-resolution ECG autonomic indices with multi-channel frontal EEG spectral power (theta/alpha ratios) to detect cognitive fatigue and memory retention states in real time.
Porting continuous wavelet transforms (CWT) and Pan-Tompkins QRS tracking from Python workstations directly onto sub-15mW ARM Cortex-M33 / TinyML edge silicon for 24/7 autonomous monitoring.
Transforming passive data collection into an active cognitive co-pilot: dynamically delivering haptic pacing, cognitive reset cues, and adaptive task throttling when parasympathetic exhaustion is detected.
Built dual ECG/PPG benchtop acquisition rig; verified 99.0% R-peak accuracy against MIT-BIH; designed 4-layer KiCad wearable PCB.
Audited 5,959 Sternberg memory trials (-1.93 bpm focus dip); validated real-world cycle variations; streaming live BLE data to custom dashboards.
Fabricate compact wearable enclosure; test motion-rejection algorithms in daily use; integrate real-time HRV feedback.