Experience

• Co-Founder & Research Scientist·Nettverk

  Sweden / Germany·Jul. 2025 – Present

  • Co-founded a voice-first AI platform for operating rooms, enabling passive clinical documentation and real-time decision support.
  • Secured SEK 1M seed investment from Johnson & Johnson AI Health; established research collaboration with Karolinska Institutet (KI).
  • Designed and led development of speech-driven ML systems for low-latency inference in noisy, multi-speaker surgical environments.
  • Built hardware-free, scalable pipelines integrating ASR, speaker understanding, and context-aware information retrieval.
• Associate Researcher·KTH Speech, Music and Hearing Lab

  Stockholm, Sweden·Mar. 2025 – Present

  • Contributed to Språkbanken Tal (CLARIN SPEECH), Sweden's national speech technology infrastructure, building scalable tools and datasets for the European Language Grid.
  • Pioneered WaveEGG, the first ML model to predict physiological voice signals (EGG) directly from acoustic input — eliminating contact-based hardware and enabling non-invasive, real-time clinical voice assessment at scale.
  • Integrated ML-based TTS evaluation into clinical-grade software, optimizing runtime performance.
• Doctoral Researcher·KTH Speech, Music and Hearing Lab

  Stockholm, Sweden·Jan. 2021 – Mar. 2025

  • Invented and led the development of VoiceMap, a hybrid DSP + ML benchmarking framework for speech synthesis, supporting large-scale multi-model comparison and voice system characterization and classification.
  • Built automated pipelines for 500+ hours of synthetic and pathological speech data, enabling real-time voice quality mapping and load-aware batch inference.
  • Collaborated with clinical and software engineers to co-design signal evaluation tools with low-latency processing and multi-platform deployment.
• TTS Engineer·TikTok / ByteDance AI Lab

  Beijing, China·Jan. 2020 – Jan. 2021

  • Engineered and productionized backend systems for TikTok/Douyin's automated voice-over tool, supporting daily generation of millions of user videos.
  • Designed an A/B testing platform for perceptual TTS quality, integrating real-time performance tracking and workload-aware model selection.
  • Reduced inference latency by 30% through model quantization and batch-serving optimization across GPU clusters under global user load.
• Research Engineer·Peking University — Linguistics Lab

  Beijing, China·Sep. 2017 – Jan. 2020

  • Architected a web-based speech data infrastructure with structured metadata, enabling query-efficient access to 1000+ hours of annotated corpus for national-scale phonetic research.
  • Engineered automated signal processing pipelines in Python for large-scale acoustic data collection, annotation, and post-processing.
• Instructional Systems Developer·Tsinghua University

  Beijing, China·Jun. 2016 – Dec. 2016

  • Built backend infrastructure for a MOOC platform serving 50,000+ learners, including automated grading pipelines, video delivery systems, and real-time analytics dashboards.