Marcus G.K. Williams.

Intel Neuromorphic Computing Lab

Senior AI Middleware Software Engineer

Research software and AI infrastructure in the Office of the CTO: language-model training and evaluation, reproducible experiments, AI operations and software for neuromorphic hardware.

• HGRN training and evaluation Pretrained and instruction-tuned 340M and 1.3B parameter HGRN language models at 90% sparsity. Evaluated checkpoints using custom tools and standard benchmarks, supporting a successful remote vehicle control demonstration on Loihi 2.
• DevKit initiative I designed and built a containerized DevKit combining our software with hardware setup scripts. The lab adopted it, reducing the work required to set up a development environment and hardware. Running the first workload went from months to about 30 minutes.
• Lava open-source engineering Led Lava’s open-source processes and GitHub setup, building CI, automated hardware testing and release pipelines that became the lab’s standard workflow for validating and releasing the Lava SDK. Contributed across the runtime, compiler and modeling software.
• Hybrid-cloud AI access Built a hybrid-cloud system combining AWS Bedrock, an AI gateway, other AI services and automated user setup to provide Claude Code and Claude Desktop access with centralized spend controls for approximately 40 NCL team members within corporate guidelines.
• Agentic & AI Workgroup Founded and lead the lab’s Agentic & AI Workgroup, sharing practices for AI-assisted engineering and agentic autonomous workflows.
• Hala Point Contributed software, workloads, testing and deployment work to Hala Point.

Intel Distributed Systems and Evaluation

Intrinsics Performance Engineer, Cloud Software Engineer & Network Software Engineer

Intrinsics Performance Engineer Feb 2020 — Oct 2021

Upstream OpenJDK performance engineering using C++, assembly and Java, including support for hardware instructions and compiler optimizations.

Cloud Software Engineer Aug 2018 — Feb 2020

Edge service orchestration in Go, with Kubernetes, Docker, gRPC, API design and microservices.

Network Software Engineer Oct 2014 — Feb 2020

Open-source networking across ONAP, OpenDaylight, OpenStack and Open vSwitch, including containerized, highly available OpenDaylight deployments in Kolla.

• Network and cloud evaluation Co-founded S3P working-group efforts across OPNFV and OpenDaylight, evaluating scale, security, stability and performance in open-source software-defined networking. Conducted scale testing with 1,800 OVSDB nodes against a single OpenDaylight controller, validating node registration and teardown for an industry performance report. Enabled containerized OpenStack deployments at scale using Open vSwitch networking.
• ONAP technical leadership Architected Service Orchestrator changes for Hardware Platform Awareness (HPA) across ONAP’s Beijing and Casablanca releases. Led cross-project integration testing supporting HPA’s inclusion in Beijing and led two engineers delivering HPA, homing and multicloud features for Casablanca. Served as a committer across APPC, SDNC, CCSDK and SO, and was selected by the technical steering committee as a Super Committer.
• Reproducible performance evaluation Wrote OpenJDK Java performance intrinsic improvements. Built an Ansible SPECjbb harness that automated bare-metal OS deployment, dependencies and test software, controlling BIOS configuration and software versions. Enabled repeatable, automated evaluation of proposed Java intrinsic performance improvements, isolating code changes from environment effects.

Intel

Software Validation Engineer

Linux network-driver validation, test automation, regression coverage and validation planning for Data Center Bridging and Fibre Channel over Ethernet.

Intel through SCIGON Solutions

Network Storage Software Validation Engineer

Network-storage driver testing, performance analysis and test automation across Windows, Linux and virtualized environments.

Propeller Heads

Consultant · Network & Systems Administration

Technology planning, network design and Windows and Linux systems administration for small businesses.

Autodidact

Independent product · Design and engineering

• Learning product Lead design and engineering of a learning application built around Faraday, an AI tutor. Connect course materials, guided reading, practice, voice, and scheduled review.
• Python practice Integrated saved Python exercises alongside reading activities, with a sandboxed browser runner and the learner’s saved work available to the tutor.

Syndicate

Independent project · Private research

• System and contribution Designed and built a platform for teams of AI agents with durable orchestration using Temporal, typed contracts, hybrid retrieval and budgeted context assembly.
• Evidence Evaluated retrieval and context compilation across 576 single-agent attempts on two models. Compared with the baseline, retrieval increased tasks solved within four attempts from 11 to 18 and from 13 to 22 out of 24. Team-scaling evaluation remains in progress.

Evolved

Independent project · Private research

• System and contribution Designed and built a modular evolutionary code-optimization system with LLM-driven search and a Firecracker VM pool for sandboxed evaluation; one 60-minute run on four GPUs evaluated 2,345 candidates.
• Evaluation Built evaluation checks that exposed misleading results, from hash-finalizer candidates that gamed the scorer to a matrix-multiplication result I withdrew after finding pseudoreplication.

Ji

Independent project · Agent research and engineering

• Agent runtime Designed and built an experimental Rust computer-agent runtime that records actions, resumes interrupted work, and runs model-directed file edits and tests. Controllers and model providers can be replaced independently.
• Controlled experiments Ran a 432-run coding study comparing reasoning replay, tool-request parsing, and repository search. The combined changes did not establish an overall improvement; independent behavioral checks determined task success.

Symphony

Independent project · Coding-agent orchestration

• Implementation and architecture Built an independent Elixir implementation from OpenAI’s original Symphony specification, then evolved both the system and its specification. Coordinates issues, lead agents, and helpers using Codex, Claude, and Pi, locally or over SSH.
• Review and delivery Added review requirements, risk checks, and CI checks tied to the commit being merged. The operator can inspect runs, pause dispatch, and rerun work; delivery-path testing and provider recovery remain in development.

M.S. Computer Science (Artificial Intelligence)

Western Governors University

In progress – expected completion February 2027.

B.S. Information Technology – Software Development

Western Governors University