Max Graf

ML Engineer · Audio & multimodal generative models · PhD, AI & Music
max.graf@live.at / website / portfolio / GitHub

Summary

ML Engineer and PhD graduate (AI & Music, C4DM at Queen Mary University of London) specialising in content-based audio retrieval, multimodal representation learning, and controllable generative audio. Hands-on with diffusion, transformers, Mamba state-space models, neural audio codecs, and low-latency C++ for edge deployment. Currently ML Engineer on the EU-funded CORPUS project, building enterprise-scale catalogue analysis and music-generation pipelines (50,000+ tracks). Proven deep-tech commercialisation track record: over £48,500 in competitive Innovate UK and impact funding, plus a competitive EuroHPC Vega allocation (~69k A100 GPU-hours).

Core Skills

Audio ML / MIR
Content-based audio retrieval (CBAR), multimodal representation learning, automated music tagging, semantic indexing, audio embeddings, source separation
Generative Audio
Diffusion, transformers, state-space models (Mamba), contrastive learning, neural audio codecs, audio tokenisation (RVQ)
Cross-Modal Retrieval
Text-to-audio alignment, audio-text embeddings (CLAP, MuLan), zero-shot classification
Acoustic Features
Timbral, rhythmic, and semantic descriptors (MFCCs, spectral centroid, chroma, onset/beat tracking, tempo, genre/mood)
ML Stack
Python, PyTorch, Lightning, scikit-learn, NumPy, pandas, TensorFlow, LibTorch
Real-Time / Creative
C++ (low-latency, edge), JUCE, Max/MSP, Pure Data, Unity, C#, XR/MR, EMG, Computer Vision
Software
Backend APIs, full-stack web, JavaScript, Node.js, Svelte, Git

Experience

ML Engineer · CORPUS
Mar 2025 – present
EU-funded music intelligence platform
  • End-to-end music metadata & analysis pipeline (automated tagging, timbral/rhythmic/semantic features, CLAP-based semantic indexing) powering the CORPUS Contribution App over a 50,000+ track catalogue.
  • Trained and deployed real-time audio-domain generation models, combining Google/DeepMind's SpectroStream codec (RVQ tokenisation) with a MusicGen-style transformer for high-quality instrumental generation.
  • Trained diffusion-based models for controlled longer-form generation (Stable Audio baseline), with inference-time steering on pitch, rhythm, and energy.
  • Built CORPUS Music Intelligence, an interactive discovery platform combining metadata, CLAP cross-modal retrieval, and diffusion generation, with full-stack integration (backend, APIs, frontend).
Researcher & Developer · Netz XR Instrument
Sep 2024 – Jan 2025
Queen Mary University of London, PhD spin-out
  • Secured Innovate UK grants to commercialise an XR musical instrument combining real-time gesture recognition and expressive sound synthesis.
  • Drove prototype iterations, user studies, and public demos as part of grant-funded commercialisation.
Researcher & Developer · WavNav / Audio Maps
May 2023 – Feb 2026
Queen Mary University of London
  • Built WavNav, an AI sample explorer applying CBAR and multimodal representation learning: CLAP embeddings, similarity metrics, dimensionality reduction for visual sample maps, and query-by-example retrieval.
  • Implemented signal-processing and feature-extraction pipelines (MFCCs, spectral centroid, chroma, onset/beat) for sample analysis and search.
  • Designed interfaces for fast exploration of large local music collections, focused on music-production workflows.
Teaching Assistant, Queen Mary University of London. Postgraduate module Interactive Digital Multimedia Techniques, workshops on real-time audio, gesture control, and XR.
Oct 2021 – Feb 2022
Software Engineer, RSIT (Vienna). Part-time full-stack and data visualisation (JavaScript, Python, Java, C++).
2016 – 2019
Earlier, software-engineering internships at ASFiNAG, Frequentis, and Roombonus.
2012 – 2016

Selected Publications

  • Graf & Barthet. Multimodal Hand Tracking for XR Musical Instruments Using Electromyography. Paper
    LNCS 2025
  • Graf & Barthet. When XR Meets AI: Integrating Interactive Machine Learning with an XR Musical Instrument. Paper
    AES AI & Musician 2024
  • Graf & Barthet. Combining Vision and EMG-Based Hand Tracking for Extended Reality Musical Instruments. Best Paper Nomination. Paper · Code
    CMMR 2023
  • Graf & Barthet. Reducing Sensing Errors in a Mixed Reality Musical Instrument. Paper
    ACM VRST 2023
  • Graf & Barthet. Mixed Reality Musical Interface: Exploring Ergonomics and Adaptive Hand Pose Recognition. Paper · Code
    NIME 2022
  • Graf, Opara & Barthet. An Audio-Driven System for Real-Time Music Visualisation. Paper · Code
    AES 150, 2021

Education

PhD, AI & Music, Queen Mary University of London (Supervisors: Mathieu Barthet & Andrew McPherson)
2020 – 2026
MSc, Sound and Music Computing, Queen Mary University of London (Distinction)
2019 – 2020
BSc, Media Informatics & Visual Computing, TU Wien
2015 – 2019

Grants & Awards

~69k
GPU-h
EuroHPC Vega HPC allocation2025
3-month distributed ML training on up to 32× NVIDIA A100 (~£60k–£100k commercial cloud equivalent).
£35,000
Innovate UK ICURe Explore2024
Commercial validation of Netz XR instrument.
£10,000
Impact Funding2024 / 2023
Netz (2024) and Audio Maps (2023).
£3,500
Innovate UK ICURe Discover2024
Early-stage commercial discovery.
Award
MIDI Innovation Award2023
For the Netz XR musical instrument.
Award
Excellent Foreign Doctoral Students2020 / 2021
Austrian national award (two years).