EV Range Estimation
An LSTM compressed to run on a $5 microcontroller
Project Overview
An end-to-end pipeline that predicts the remaining range of a Keke Maruwa electric tricycle from live telemetry, then runs that prediction on the vehicle itself rather than in the cloud. Range is a time-series problem: a tricycle driven hard for the last minute has very different remaining range than one driven smoothly, even at identical state of charge and battery temperature. The model therefore reads a 60-second rolling window of ten sensor features rather than an instantaneous snapshot.
Client
Self-initiated
Role
Sole engineer (modelling, quantization, firmware)
Completed
July 2026
Duration
Research and build
Technologies Used
Frontend
Backend
Deployment
Other Tools
Related Projects
Artemis
A smart-home agent that turns speech into tool calls and controls physical devices, gated by an explicit reasoning trace and user approval before anything happens.
Attention-Based Malaria & TB Screening
CBAM attention written from scratch and injected into five CNN architectures, compared with McNemar significance testing rather than raw accuracy deltas.