Will Fellhoelter

Eight years across the stack: data, application logic, deploys, the K8s underneath, the alerting on top. I usually end up in the gap between what a client needs and what actually ships, and closing it starts with listening, so what gets built is what was needed. Lately that work points at agentic AI: MCP servers, plugin libraries, and the knowledge bases agents work from.

Download resume

Will Fellhoelter laughing, in sunglasses, against a blue sky — Currently01Principal Software Engineer at Accuris. Agentic AI development, Operations, K8s.02Building Sauce. An agentic operating loop for Obsidian, open source.

01Experience

Eight years, seven roles.

Full resume, PDF

Mar 2026 – NowAccurisPrincipal Software EngineerMCP Mesh, observability, multi-region K8s
- With the platform team, built and launched the MCP Mesh on AgentGateway: one governed entry point to 250+ tools for engineering, product, and cross-functional teams, with SSO and centralized observability
- Designed and shipped production MCP servers that give AI agents natural-language access to internal systems, including our own New Relic and PagerDuty servers in place of paid vendor connectors
- Built the incident plugin that takes a PagerDuty alert to a root cause in New Relic and documents it in the team’s Obsidian vault in the same pass
- Grew the AI plugin library into how teams use AI day to day, turning one-off prompts into repeatable workflows
- With the platform team, rolled out the New Relic K8s operator so 400+ deployed services get consistent APM injection with no manual work from developers, standardizing naming and troubleshooting across the org
- Operate 100+ microservices on K8s across 6 regions; led the multi-region buildout of compute and data tiers
- Help map out and drive the standardization of deployment and CI/CD across 1,000+ repositories with templated K8s
- Build and run CDC pipelines from PostgreSQL into Databricks for mission-critical data
Apr 2025 – Mar 2026AccurisSenior Software EngineerGitHub migration, first internal MCP servers
- Led the move of 1,000+ repositories to GitHub and consolidated CI onto GitHub Actions with reusable workflows
- Built the first internal MCP servers and the original AI plugin library
Aug 2024 – Apr 2025formlFull Stack Engineer, Senior from Jan 2025First engineering hire, on-prem deploys
- First engineering hire; worked directly with the founders and enterprise clients to scope requirements, set quarterly roadmaps, and ship the platform end to end (Python, Angular, AWS, PostgreSQL)
- Built one-click on-prem deployment for enterprise clients, cutting onboarding from 3 days to 4 hours and unblocking deals that required customer-hosted installs
- Led the API and architecture work that carried the platform through 2x user growth in 2 months; promoted to Senior
May 2023 – Aug 2024Project CanaryDevOps EngineerEphemeral environments, RAG, SOC 2
- Built a self-service tool that lets developers and QA deploy an isolated test environment from any branch in one click, with selectable data sources and automatic cleanup
- Built a RAG search app on AWS Bedrock and vector databases for querying internal documents across departments
- Ran observability for 10+ critical services (99.99% uptime) and moved it to New Relic, cutting monitoring costs 14%
- Fixed bottlenecks in high-throughput time-series PostgreSQL databases, improving query performance 23%
- Drove the security work behind passing SOC 2 Type I and Type II, automating controls and cutting audit prep ~30%
Aug 2021 – May 2023LendflowDevSecOps EngineerOpenSearch logging, WAF, on-call
- Centralized logging on OpenSearch, cutting the time to pinpoint production issues from ~3 hours to under 10 minutes
- Built ephemeral dev and QA environments to replace always-on ones, cutting environment costs 32%
- Put AWS WAF and CloudFront in front of the platform, reducing security incidents 60%
- On-call engineer for a platform serving 20K+ users; cut average ticket resolution time 25%
Mar 2020 – Aug 2021MCG HealthCloud / DevOps EngineerOn-prem to Azure, Chef to Ansible
- Led the migration from on-prem infrastructure to Azure for 3+ teams
- Built modular Terraform components that cut new environment setup from ~2 days to 1 hour
- Moved configuration management for 100+ Windows servers from Chef to Ansible, with Datadog monitoring
Oct 2018 – Mar 2020Cerner CorporationSystem EngineerAnsible automation, AWX portal
- Wrote Ansible playbooks for server configuration and troubleshooting, saving the team 40+ hours a month
- Built an AWX-based portal for scheduling data restores and spinning up ephemeral test environments
- Taught biweekly Linux fundamentals sessions for new hires through Cerner’s Tech Academy
2018Wichita State UniversityManagement Information Systems

02Selected work

Work I’d point you to.

Agent tooling for a whole org, observability for hundreds of services, installs on someone else’s hardware. Built with platform, product, and engineering teams along the way.

Accuris · 2025–2026 · With the platform team

MCP Mesh

Governed agent tooling for the whole org. AgentGateway, the open-source gateway for MCP and agent traffic, runs on K8s and federates 250+ tools behind one endpoint, with SSO and central observability. Built the first internal MCP servers and the plugin library, then, with the platform team, the mesh. Our own servers for New Relic and PagerDuty live inside it, written to avoid paying for vendor connectors.

agentgateway.dev · the open-source project

Every teamEngineering, product, cross-functional

Plugin libraryRepeatable workflows, not one-off prompts

AgentGatewayOn K8s: one endpoint, SSO, central observability

250+ toolsFederated MCP servers, including our own New Relic and PagerDuty

Accuris · 2026 · Runs on the mesh

From a page to a documented root cause

Point the plugin at a PagerDuty alert. It works through the team’s skills and our own MCP servers, traces the likely root cause in New Relic, and files the write-up in the team’s Obsidian vault. Troubleshooting and documentation happen in the same pass.

PagerDuty alertThe page, linked straight from the plugin

Agentic pluginTeam skills plus our own MCP servers

Likely root causeFrom New Relic, through our own MCP server

Obsidian vaultWrite-up filed for the team, same pass

Accuris · With the platform team
The same observability for every service
Configured the New Relic K8s operator with the platform team so every service deployed across the org, 400+ of them, gets the same APM injection automatically. No manual setup for developers, consistent naming, and one way to troubleshoot across the org.
400+
Services across the org, instrumented the same way
forml · 2024–2025
On-prem in four hours
First engineering hire at an AI startup. Worked with the founders and enterprise clients to understand what they actually needed, then built one-click on-prem deployment so the deals that required customer-hosted installs could close.
3 daysto4 hours
Enterprise onboarding time
Cerner · Lendflow · Project Canary
Ephemeral environments, three times
Self-service test environments with data seeding and automatic cleanup, built for three different orgs: an AWX portal at Cerner, ephemeral infrastructure for dev and QA at Lendflow, a UI with branch and data-source selection at Project Canary. The problem I keep getting handed.
−32%
Environment costs at Lendflow

03Personal projects

Sauce

Open source · 2026– · MIT

An agentic operating loop for Obsidian. Notes become node-based work graphs that agents like Claude Code and Codex execute in isolated workers, with scheduling, retries, handoffs between agents, and a persistent task store. Started as a loop-based forward deployment mechanism, rebuilt around graphs. Versioned vault platform shipped via Homebrew, MIT.

How a run works

Graph noteAgent, judge, shell and human nodes in an Obsidian note

Dispatchersauce run fires edges, with retry budgets

WorkersClaude Code and Codex, each in its own git worktree

Back into the vaultAn append-only ledger and a Runs section in the note

$ brew tap willfell/sauce$ brew install willfell/sauce/sauce

github.com/willfell/sauce

Obsidian on a phone showing the epic note Delivery Coordinator Rail Repairs: three slice cards, OPS-1, OPS-2b and OPS-3c, each marked done, and a rollup reading 3 deployed — An epic and its slices on a phone. Card status is projected from the coordinator’s ledger, not typed by hand.

A dependency graph in Obsidian with slice cards PERF-9a in progress and PERF-10a done, and a dashed placeholder node standing in for work in another epic — A wider dependency graph. The dashed node stands in for a dependency that lives in another epic.

04GitHub

9,019contributions in the last year

Most since May are Sauce: agents do the work in their own git worktrees, and I steer and merge what lands.

github.com/willfell

Oct

Nov

Dec

Jan

Feb

Mar

Apr

May

Jun

Jul

Aug

Sep

Last 12 months on GitHub

willfell/sauce
Agentic operating loop for Obsidian: node-based agent execution, versioned vault platform, shipped via Homebrew
JavaScriptMIT
willfell/will-fell
This site. Next.js and Terraform on S3 and CloudFront, deployed by GitHub Actions on every merge.
TypeScriptTerraform