BioSmartSense: IoT Electrochemical AI Platform

Portable IoT electrochemical impedance biosensor with machine learning for real-time environmental water quality monitoring and intelligent healthcare diagnostics.

Comment

Inspiration

Water contamination, heavy metal pollution, and delayed physiological biomarker detection threaten millions of vulnerable lives globally. While electrochemical sensing (potentiostats and Electrochemical Impedance Spectroscopy / EIS) offers unmatched sensitivity and specificity, traditional laboratory instruments are bulky, benchtop devices costing upwards of $5,000 to $15,000 USD. This economic and form-factor barrier leaves underserved communities and mobile field researchers without access to decentralized, real-time chemical diagnostic tools.

We created BioSmartSense to break this barrier: integrating custom-engineered, ultra-compact IoT electrochemical hardware with edge machine learning algorithms to enable continuous, accessible, and high-accuracy point-of-care environmental water testing and personalized healthcare sensing.

What it does

BioSmartSense is an open-architecture, closed-loop IoT biosensing intelligence system:

  1. Precision Edge Hardware: A custom 4-layer High-Density Interconnect (HDI) PCB built around the Analog Devices AD5940 high-precision analog front end (AFE). It natively interfaces with standard 3-electrode gold/carbon Screen-Printed Electrodes (SPE) to conduct Electrochemical Impedance Spectroscopy (EIS, 0.016 Hz to 200 kHz) and Cyclic Voltammetry (CV).
  2. Machine Learning Signal Interpreter: Automatically filters non-faradaic background capacitance, baseline drifts, and high-frequency parasitic noise. It extracts Randles equivalent circuit parameters ($R_{ct}$, double-layer capacitance $C_{dl}$, and Warburg diffusion $Z_w$) to classify analyte concentrations and heavy metal ion presence (e.g., $Pb^{2+}$, $Cd^{2+}$) within seconds.
  3. Telemetry & Real-Time Cloud Dashboard: Live telemetry streams real-time Nyquist and Bode impedance plots to a cross-platform web console, offering instant hazard alerts and field diagnostics for environmental social good.

How we built it

  • Hardware Design (KiCad 8): Engineered an ultra-low-noise 4-layer PCB stackup with dedicated analog ground planes, active shielding guard rings for picoamp-level ($pA$) leakage suppression, ultra-low-ESR power filtering, and standardized SPE card-edge sockets.
  • Embedded Firmware: Configured SPI high-speed register sequencing, complex-number digital discrete Fourier transform (DFT) hardware accelerators, and excitation AC sinusoidal synthesis on the AD5940.
  • Machine Learning & Signal Processing: Developed a Python-based processing pipeline leveraging NumPy, SciPy, and automated regression models for rapid spectrum deconvolution and peak identification.
  • Software Interface: Designed an interactive telemetry and diagnostic dashboard using Streamlit and FastAPI for real-time visualization, calibration, and automated report generation.

Challenges we ran into

  • Picoamp-Level Noise Mitigation: Measuring high-impedance biological and environmental samples on a handheld board introduces significant ambient electromagnetic interference (EMI). We solved this by implementing differential analog layout routing, ground guard traces, and software moving-average digital filters.
  • Electrochemical Non-Linearities: Real-world water and biological matrices exhibit complex electrical double layers that distort classic linear models. We trained a lightweight multi-feature regression model to map multi-frequency phase-angle shifts directly to actual concentration levels.

Accomplishments that we're proud of

  • Successfully engineered and validated the complete production-grade hardware package in KiCad, including schematic, 4-layer PCB layout with DRC pass, and JLCPCB-ready manufacturing files.
  • Built a working multi-frequency EIS feature extraction pipeline achieving over 95% classification accuracy on test chemical benchmarks.
  • Kept the total hardware bill of materials (BOM) cost under $35 USD—making high-precision electrochemistry accessible at 1% of the cost of commercial laboratory workstations.

What we learned

  • How to bridge the gap between physical analog electronics and modern data science / machine learning.
  • The immense potential of democratization in hardware: combining open-source EDA tools with machine learning to solve real-world environmental health challenges.

What's next for BioSmartSense

  • Fabricating the Rev-A2 hardware with onboard Bluetooth Low Energy (BLE) and LiPo battery management for standalone field deployment.
  • Integrating functional hydrogel dressings for wearable, non-invasive wound exudate monitoring and smart healthcare delivery.

Built With

Share this project:

Updates

Submission history