Abhay builds embedded firmware, high-voltage boards, CAN networks, race cars and much more embedded firmware high-voltage boards CAN networks race cars and much more embedded firmware

• 

  Battery management system

  600 V

  Ojas Racing, 2024–2026

  Custom BMS for a 3P140S Formula Student accumulator. STM32H7 master with LTC6804 cell monitors over isoSPI, DMA-based voltage acquisition, CAN telemetry, cell balancing, fault detection and SoC estimation. I did the schematics, multi-layer PCBs, firmware and bring-up.

  STM32H7LTC6804isoSPICANAltium
• 

  High voltage fault detection board

  HV

  Ojas Racing

  A high-voltage board for the Ojas car with isolated high-voltage and low-voltage sides, built to detect faults.

  High voltageIsolationPCB design
• 

  I-Align v2.0

  35 × 45 mm

  Accepted for panel discussion, VRSI National Conference

  A wearable that tracks orientation for patients recovering from retinal surgery. Compact board with an ESP32-C3, MPU6050 and EEPROM; Kalman-filter orientation tracking, on-board logging and deep-sleep power saving.

  ESP32-C3MPU6050Kalman filterKiCad

• CAN bus intrusion detection

  67 ms

  Paper accepted at iSENSE, IIIT Kottayam

  A GRU-based intrusion detector that runs on the microcontroller itself. Two ESP32 nodes sit on a 500 kbps CAN bus: one captures frames in listen-only mode, the other runs a 16-unit GRU (TFLite Micro) over 20-frame windows. It classifies DoS, fuzzing and spoofing traffic with 99.1% accuracy and 67 ms end-to-end latency, in 82 KB of flash and 29 KB of RAM, ahead of 1D-CNN and threshold baselines. It is moving to an STM32H7 next, with far more functionality.

  ESP32TWAI / CANFreeRTOSTFLite MicroPython
• 

  EKF state-of-charge estimator

  <2% error

  SMART Lab, IIT Madras, 2025–2026

  An extended Kalman filter in MATLAB that estimates a lithium-ion cell’s state of charge, validated against lab charge–discharge cycles with under 2% error.

  MATLABEKFBattery testing

• LFP degradation and equivalent-circuit model

  2RC

  UD Trucks, 2026

  A physics-informed model of LFP capacity fade and resistance growth, coupled to a 2RC equivalent circuit parameterised over temperature, SoC, cycle count and age for SoH estimation.

  MATLABSimulinkPyBaMMECM

• Icare-ML

  2.90 mmHg

  GitHub

  Predicts Goldmann applanation tonometry (GAT) eye pressure from central corneal thickness and iCare tonometer readings, trained on about 6,600 clinical eye records. A tuned XGBoost model brings the test error to 2.90 mmHg MAE (R² 0.73), down from 3.21 mmHg when the iCare reading is taken as-is.

  PythonXGBoostscikit-learnRegression

  View on GitHub

Experience

Education