Portrait of Markus Plandowski

MARKUS PLANDOWSKI

I build compact neural networks for time series and make them fast for real-world systems.

Independent Research — Battery State Estimation
06/2025 - Present | Wertingen, Germany

Self-directed R&D on compact neural battery surrogate models, spanning battery-data infrastructure, reproducible training, and long-horizon validation.

  • Built reusable Polars data infrastructure for heterogeneous cycling data, reducing new-dataset onboarding to schema mapping while preserving reproducible preprocessing, streaming access, and safeguards against silent dataset-version mismatches.
  • Developed scalable visualization and LTTB downsampling tools for inspecting dataset quality and long-horizon rollouts across millions of samples.
  • Created reproducible Docker/config-driven training and HPO workflows using Optuna/W&B, checkpoint comparison, and experiment tracking.
  • Exposed profile memorization in repetitive cycling data, shifting validation toward randomized profiles, OOD discharge extensions, and broader real-data diversity.
  • Prototyped compact PyTorch state-space and attention-based surrogate models for metadata- and control-conditioned battery dynamics.

Battery Machine Learning Consulting 11/2024 - 05/2025 | Chemix Inc | Remote / Sunnyvale, USA

Contributed to a Silicon Valley battery startup's ML stack in a fast-moving production codebase.

  • Introduced reproducible HPO workflows using SMAC3, Dask, and W&B, identifying a model with 10× fewer parameters and better validation loss than the baseline while preserving checkpoint compatibility.
  • Reproduced standardized cycling protocols in PyBaMM to generate long-horizon synthetic data aligned with physical test-cycler results for model validation.
  • Operated as a remote external contractor, aligning scope with the in-house team, contributing design reviews, and supporting async implementation cycles across time zones.

Battery Machine Learning Open Source Cell-Li-Gent 04/2024 - 10/2024 | Wertingen, Germany

Scaled thesis-stage SoC estimation into a multi-output battery model for terminal voltage, surface/core temperature, and OCV on synthetic SPMe data.

  • Refined a vanilla GPT-style baseline into a roughly 16M-parameter domain-specific Transformer through HPO, enabling practical autoregressive rollouts.
  • Created a synthetic PyBaMM/SPMe pipeline with randomized profiles, datasheet limits, and thermal signals.

Solo Motorbike Travel | 10/2023 - 03/2024 | Vietnam

Master Thesis and Research Internship 06/2022 - 09/2023 | Fraunhofer ISE | Freiburg i.Br., Germany

Self-proposed thesis on data-driven SoC estimation across NMC/64Ah and LFP/2.6Ah cells using synthetic ECM data, transfer learning, and meta-learning.

  • Benchmarked tree-based, recurrent, convolutional, state-space, and Transformer models under matched HPO budgets, finding Transformers strongest among neural models for transfer while tuned ECMs remained the accuracy baseline.
  • Preceding internship: recurrent architectures including stacked LSTM, Grid-LSTM, and bidirectional Grid-LSTM for battery state estimation.

Embedded Machine Learning Internship 06/2021 - 05/2022 | Telocate GmbH | Freiburg i.Br., Germany

End-to-end AI smart-glove prototype for real-time object-presence detection.

  • Designed a four-channel acquisition PCB and implemented the VHDL FPGA data path, converting conditioned parallel SPI sensor streams into a real-time-capable UART interface using Altium, LTspice, and MATLAB.
  • Benchmarked multivariate time-series classifiers, including MultiRocket and InceptionTime, against an MLP baseline for binary object-presence detection, achieving near-perfect accuracy in prototype validation.

M.Sc. Embedded Systems Engineering 11/2020 - 09/2023 | University of Freiburg | Freiburg i.Br., Germany

Hands-on embedded-systems, sensing, modelling, and AI coursework.

Work and Travel | 10/2018 - 10/2020 | Asia-Pacific and Australia (licensed electrical work)

Battery Engineer, Full-Time 04/2017 - 09/2018 | Mahle GmbH | Stuttgart, Germany

Owned electrical-interface design for a high-voltage immersion-cooled 800V/750kW battery-pack prototype, coordinating across internal stakeholders, suppliers, and OEM reviews.

  • Coordinated electrical integration from module prototype to full-pack design, covering BMS sensing, connectors, protection, and HV safety under automotive functional-safety constraints.
  • Named inventor on patent publications US20200153060 and US20200119415 related to the battery system work.
  • Developed transient electro-thermal busbar simulations to evaluate hotspot formation during rapid design iterations.

Bachelor Thesis and Hardware Internship 04/2016 - 03/2017 | Mahle GmbH | Stuttgart, Germany

Modeled a 14V vehicle electrical network to evaluate integration of a thermoelectric waste-heat recovery system and bidirectional DC/DC converter.

  • Created and validated onboard network simulations covering TEG, DC/DC, alternator, battery, and loads, and assessed recuperation states (Simulink/PLECS).
  • Supported hardware bring-up through soldering, driver and sensor testing, SPICE fault analysis, and thermal test-bench measurements.

B.Eng. Mechatronics 10/2012 - 03/2017 | Ulm University of Applied Sciences | Ulm, Germany

Developed a stronger focus on embedded systems and simulation.

Electrician and Mechatronics Apprenticeship 08/2008 - 09/2015 | Osram AG | Augsburg, Germany

Returned during university for production-line electrical work, covering switchboard wiring, high-power cabling, fault tracing, and small-team coordination under deadline pressure.