Summary
Overview
Work History
Education
Skills
Languages
Timeline
Generic

Magnus Hyttsten

Summary

Dynamic technology leader with a proven track record at Google, specializing in GenAI system evaluation and LLM training. Co-founder and CTO of DigitalRoute, I excel in entrepreneurship and team management, driving innovation in AI/ML compilers and frameworks. Recognized for enhancing model capabilities and delivering impactful solutions in high-stakes environments.

Overview

25
25
years of professional experience

Work History

GenAI Evaluation Systems

Google
Mountain View, CA
05.2024 - Current
  • Constructed an evaluation methodology to critically evaluate GenAI system quality, differentiating between stochastic/deterministic evaluation, and automated, human, and LLM as a judge evaluation.
  • Defined evaluation taxonomy based on job-to-be-done frameworks, including LLM personas like job-description, factual accuracy, skill (reasoning) accuracy.
  • Deployed the system for Agentic LLM Chat and Coding applications to provide statistical certainty on quality improvement/regression based on system variations including: model, prompt engineering, factual grounding, Agentic setup/tool usage.
  • Worked across business, development, and production organizations to align and execute launches

LLM Model Training & Deployment

Google
Mountain View, CA
04.2023 - 06.2024
  • Led engineering team enhancing foundational Gemini models with continued pre-training (e.g. FIM, TreeTrim) and supervised fine-tuning (SFT/IT, RLHF/RL) for Google Cloud purposes.
  • Worked across models and teams: balancing requirements related to knowledge needs (datasets), model sizes (parameter counts), serving/latency costs (hardware requirements), and application specificity versus foundational model capability.
  • First-hand professional experience of AI industry evolution of model capabilities, fine-tuning, grounding/RAG architectures, context-learning, and Agentic workflows
  • Collaborated with team to design worldwide deployment architecture, considering hardware cost, latency, and availability based on application needs

AI/ML Compilers & Runtimes for GPUs, TPUs, XPUs

Google
Mountain View
02.2022 - 03.2023
  • Developing compiler stack for machine learning accelerators such as GPUs, TPUs, XPUs, and associated runtimes.
  • Launched OpenXLA (https://github.com/openxla) and StableHLO, based on MLIR under LLVM. Aligned hardware ecosystem: Nvidia, ARM, AMD, Intel, Graphcore, Cerebras, AWS, etc, under OpenXLA.
  • Worked with various optimization techniques all the way from low-level kernels (e.g. Cuda/Triton), up to model/data distribution (single-box, pod, infinite scale)

TensorFlow Engineering

Google
Mountain View, CA
01.2017 - 02.2021
  • TensorFlow (tensorflow.org) engineering manager, looking after open-sourcing & community efforts
  • Team assessed usability of products & APIs, create sample models & code, write technical documentation (tensorflow.org) and create educational material.including packaging, documentation, sample writing, events, community efforts, etc
  • Worked with ~20 TensorFlow products (TF Lite, TFX, TF Probability, TF Quantum, Robotics, etc

Developer Relations Lead

Google
Mountain View
09.2013 - 01.2017
  • Built and managed the Developer Relations team for Firebase (firebase.google.com). Firebase provides 18 Cloud-based products to help developers build better apps on the Android, iOS, and Web platforms.
  • Led the team to assess API usability/design, develop code samples, technical documentation (firebase.google.com/docs), and educational/marketing material.

Co-Founder & CTO: DigitalRoute

DigitalRoute
Stockholm
01.2000 - 09.2013
  • Co-founder and CTO at DigitalRoute (www.digitalroute.com) between 2000-2013, being a key founding member building its technology, sourcing its capital, and aquiring customer base from scratch.
  • DigitalRoute has since its inception revolutionalized data integration for the telecommunications' industry and has year after year received recognitions and awards for its technology, revenue growth, and satisfied customers.
  • DigitalRoute grew from inception with venture capital funding to profitability, with >300 customers worldwide, and hundreds of people employed, and was acquired in 2024
  • Established DigitalRoute's (www.digitalroute.com) office in Atlanta, GA, USA. Was the engineering lead to successfully design & deploy a mission critical data integration platform for one of the largest telecommunication providers in the US, processing multi-billion records per day.

Education

Mathematics And Computer Science

University of Uppsala
Uppsala

Elementary Law

Lund University
Lund

Skills

  • Entrepreneurship & Leadership: Co-Founder & CTO of DigitalRoutecom from inception, capital, customer, company buildout, to acquisition
  • GenAI System Creation: Agentic frameworks, grounding/RAG, context learning, prompt engineering, system evaluation
  • LLM Training and Fine-tuning: Pre-training, SFT/IT, LoRA Model/data distributed training
  • AI/ML Accelerator Compilers: GPUs, TPUs, XPUs Cuda/Triton, OpenXLAorg, MLIR
  • AI/ML Frameworks: TensorFlow (part of original engineering team), Jax, PyTorch
  • Software Engineer, Technical Lead, and Manager: Managed teams up to 40 members Was part of C standardization organization

Languages

English
Full Professional
Swedish
Native/ Bilingual
Spanish
Professional

Timeline

GenAI Evaluation Systems

Google
05.2024 - Current

LLM Model Training & Deployment

Google
04.2023 - 06.2024

AI/ML Compilers & Runtimes for GPUs, TPUs, XPUs

Google
02.2022 - 03.2023

TensorFlow Engineering

Google
01.2017 - 02.2021

Developer Relations Lead

Google
09.2013 - 01.2017

Co-Founder & CTO: DigitalRoute

DigitalRoute
01.2000 - 09.2013

Mathematics And Computer Science

University of Uppsala

Elementary Law

Lund University
Magnus Hyttsten