Time-series forecasting on STM32
Embedded AI Engineer (project-based) · Tibcon Energy Technologies Inc. · October 2025 – April 2026
An end-to-end AI pipeline for time-series forecasting: from training to the microcontroller.
What I did
- Trained LSTM models in Keras, optimised them with X-CUBE-AI and deployed them on STM32.
- Board: STM32F429I-DISCOVERY (STM32F429ZI, 180 MHz, 2 MB Flash, 256 KB RAM). The model uses 533 KiB of Flash and 126 KiB of RAM, so the originally targeted STM32F446ZE was not enough.
- Developed the embedded GUI in TouchGFX to display real-time inference results; the project was delivered in fully working condition.
- Presented the resulting solution within the scope of a TÜBİTAK programme.
Tools
Keras · LSTM · STM32F429I-DISCOVERY · X-CUBE-AI · TouchGFX · C
