Looking for a job

Michael Williams

@michaelwilliams4

Senior Machine Learning Engineer at ElevenLabs who co-designed its multilingual TTS architecture and 3x faster inference path.

United States

Message

At ElevenLabs, I co-designed a text-to-speech architecture with a decoder-only backbone, RVQ codec head, and audio-tag conditioning across more than 70 languages. I also developed its TensorRT inference path, achieving a 3x end-to-end speedup at parity quality.

At AssemblyAI, I designed streaming speech-recognition models and led acoustic-encoder pretraining. At Accenture, I led RAG engagements and built an agentic-AI accelerator; earlier, at Meta, I contributed to PyTorch speech-synthesis examples and multi-GPU training infrastructure.

Experience

Work history, roles, and key accomplishments

Senior Machine Learning Engineer

ElevenLabs

Jun 2025 - Jun 2026 (1 year)

Co-designed TTS architecture with decoder-only autoregressive backbone and RVQ codec head. Engineered distributed training on H100 clusters and developed real-time audio generation pipeline with sub-150 ms latency.

PyTorch Distributed Data Parallel Fsdp TensorRT FP8 INT8 WebSockets WebRTC IPA SSML F0 Conditioning GQA RVQ Codecs NCCL Bf16

Senior Machine Learning Engineer

Accenture

Jul 2021 - May 2025 (3 years 10 months)

Led RAG engagements and architected agentic AI accelerator on LangChain and LangGraph. Trained domain-adapted ASR models and fine-tuned LLMs with QLoRA, while mentoring junior engineers.

Lang Chain LangGraph RAG QLoRA ASR LLM Evaluation PII Scrubbing Sagemaker Computer Vision Mentorship

Senior Machine Learning Engineer

AssemblyAI

Jul 2019 - Jun 2021 (1 year 11 months)

Designed streaming speech-recognition models and led self-supervised pretraining. Owned model optimization and AWS inference platform, achieving GPU utilization targets.

ASR Transformers Transducers Self Supervised Learning Quantization AWS Amazon EKS Sagemaker S3 Amazon SQS Kinesis Lambda Cloudwatch Xray

Deep Learning Engineer

Machine Learning Research Intern

Google

Jun 2017 - Sep 2017 (3 months)

Implemented audio preprocessing and dataloader code for TTS experiments. Prototyped attention-based encoder-decoder variants in TensorFlow and built visualization utilities.

TensorFlow Audio Preprocessing Data Loader Attention Mechanisms Visualization