Isak Karlsson

Software Engineer

Builder first — right tool for the job, not the other way around.

cv@isak.tech LinkedIn Tacoma, WA

Summary

A builder, first and foremost. I start from the problem and the people it serves, keep the goal in view, and reach for the right tool for each job rather than forcing a favorite — adapting as the work demands. That instinct sits on top of 7+ years of industry experience across AWS and Raytheon, 15+ years of self-directed programming and design, and formal training in computer science and mathematics at UC Berkeley. Work spans systems and distributed-systems engineering, systems integration, and applied machine learning and reinforcement learning. Previously held a DoD Secret clearance (2018–2023).

Experience

Software Engineer II · Amazon Web Services (AWS) Jan 2026 – present
  • Building an LLM-based autonomous institutional-knowledge agent that recovers knowledge lost to employee churn and turnover. It crawls internal infrastructure, pipelines, ETLs, code bases, and documentation systems (wiki, Quip, SharePoint) and aggregates them into a curated, continuously maintained knowledge base.
  • Designed the self-healing curation system: event- and schedule-driven automation prunes, cleans, merges, and expands the knowledge base as new events arrive, with a "curator" filtering inputs and several learning systems maintaining accuracy.
System Development Engineer II · Amazon Web Services (AWS), Seattle, WA Oct 2022 – Feb 2026
  • Built infrastructure and systems for supply-chain planning.
  • Worked across application security and large-scale distributed systems.
  • In the role's final year, initiated the LLM-based institutional-knowledge agent described above — now continuing in the current role.
Systems Engineer · Raytheon Missiles & Defense, Tucson, AZ Sep 2018 – Oct 2022

Lead engineer across AI/ML efforts: object detection and classification in SAR and IR imagery, edge-processor development and integration (NVIDIA Tegra), reinforcement learning for large-scale swarm autonomy, and learning-to-optimize for combinatorial problems in distributed systems. Instructor for Raytheon's Advanced Technical Education Program (reinforcement learning).

Research Assistant · Dr. Peter Cheeseman, Berkeley, CA 2015 – 2017
Teaching Assistant · University of California, Berkeley 2016

Selected Work & Research

Edge Computing 2020 – 2022

Development and integration on modern edge processors (NVIDIA Tegra); research on non-Von-Neumann architectures for edge processing.

Reinforcement Learning 2019 – 2022

Instructor for Raytheon's Advanced Technical Education Program; OpenAI Gym environment lead developer and agent researcher.

Image Classification 2018 – 2022

Object detection and classification in SAR and IR imagery.

Learning to Optimize 2020 – 2022

Learning to solve combinatorial optimization problems in large distributed systems.

Distributed Learning 2020 – 2021

Reinforcement learning applied across swarms of 100s or more units.

Swarm Autonomy 2019 – 2021

Partner effort with Caltech CAST on autonomous aerial swarm systems.

Education

University of California, Berkeley 2014 – 2017

B.A. in Computer Science and Mathematics. GPA 3.542.

Saddleback Community College 2014

A.A. in Mathematics, Magna Cum Laude.

Skills

Languages

C / C++Advanced
PythonAdvanced
HTML / CSSAdvanced
C#Intermediate
JavaIntermediate
BashIntermediate
SQLIntermediate
MIPS / RISC-VBeginner

Tools & Frameworks

AWSAdvanced
TensorFlowAdvanced
DaskAdvanced
PyTorchIntermediate
LinuxIntermediate
DockerIntermediate
SLURMIntermediate
OpenCV / scikit-imageIntermediate
scikit-learnIntermediate
CUDABeginner

Notable Projects

Volumetric Raytracing 2017

UC Berkeley senior design project (with Noah Pitts). Recreated crepuscular rays via volumetric scattering, accelerated with NVIDIA CUDA and OptiX. Recognized as a top-10 project and showcased at semester's end. Demo.

Mastering Threes 2015 – 2017

Guided research under Prof. Stuart Russell. Monte Carlo Tree Search guided by an Inception-Net-based DQN. Demo.

3D-Printed Hexapod 2017

Designed in Fusion 360 and 3D-printed; autonomously driven by a Raspberry Pi, programmed in Python.

WiiRemote Rover 2012 – 2013

Wireless 4-wheel rover with an arm, controlled by a Wii Remote with actions previewed via a game engine. Renderer · Wii Remote.

Timbre as a Continuous Function 2014

Built a continuous function in Max/MSP interpolating between timbres — piano, violin, and flute.

Multiplayer Networking Chapter 2012

Author, networking chapter of C4 Engine: 3D Game Editing & Programming (1st ed.), ed. Dr. Mohd Fairuz Shiratuddin & Dr. Mark Fruchtaman.