Carl Thomé
Stockholm, Stockholm County, Sweden
Summary
🤩
Rockstar🎓
Top SchoolCarl Thomé helps advanced ML teams turn research into production-ready products, bringing 11 years of applied ML experience underpinned by a theoretical computer science education from KTH. He combines deep learning operations, cloud engineering, and DSP expertise—particularly in music and audio ML—to lead end-to-end projects from prototyping through deployment and monitoring. A practical engineer with strong software sensibilities, he has contributed performance and low-precision support to Keras and advanced ConvLSTM implementations for TensorFlow, as well as signal-processing work in librosa. Comfortable handling large-scale datasets and feature preprocessing for GenAI models, he also mentors teams and maintains a long history of open-source and scientific community involvement. An intriguing detail: his background in music technology informs a specialty in self-supervised audio representation learning and differentiable programming that bridges research and productization.
11 years of coding experience
Civilingenjör, Computer Science, Civilingenjör, Computer Science at KTH Royal Institute of Technology
English, Swedish
Stackoverflow
2,722reputation
114kreached
15answers
25questions
python
top-5%
tensorflow
top-5%
neural-network
top-5%
keras
top-5%
Github Skills (31)
scipy10
convolutional-neural-networks10
python10
signal-processing10
machine-learning10
recurrent-neural-networks10
audio-analysis10
digital-signal-processing10
midi10
midi-instrument10
keras10
lstm10
deep-learning10
tensorflow10
neural-network10
Programming languages (22)
Github contributions (5)
carlthome/tensorflow-convlstm-cell
Dec 2016 - Jan 2022
A ConvLSTM cell with layer normalization and peepholes for TensorFlow's RNN API.
Role in this project:
ML Engineer
ML EngineerContributions:70 commits, 1 PR, 67 pushes in 5 years 1 month
Contributions summary:Carl primarily contributed to the development of a ConvLSTM cell for TensorFlow's RNN API. They implemented the core convolutional LSTM architecture, including the necessary mathematical operations for gated recurrent units. The user's work involved several iterations, refactoring and optimizing the cell, including adding layer normalization, and enabling different data formats.
apirnnnormalizationtensorflowlayer-normalization
librosa/librosa
Sep 2016 - Aug 2019
Python library for audio and music analysis
Role in this project:Data Scientist
Data ScientistContributions:33 commits, 14 PRs, 123 comments in 2 years 11 months
Contributions summary:Carl primarily focused on feature engineering and refinement within the librosa library for audio and music analysis. Their contributions involved modifying the `rmse` function, adding new functionality for RMS energy calculation, and refactoring code to prefer the time-frequency domain. They also improved test coverage by adding consistency checks. The user's work demonstrates a deep understanding of signal processing techniques and their practical application in the domain of music information retrieval.
python-librarydtwpythonlibrosaaudio
Find and Hire Top DevelopersWe’ve analyzed the programming source code of over 60 million software developers on GitHub and scored them by 50,000 skills. Sign-up on Prog,AI to search for software developers.