AI SOFTWARE ENGINEERX-ENERGY

I build AI systems that have to work in the real world.

At X-energy, I build enterprise software at the intersection of advanced nuclear and artificial intelligence. Before code, I learned to deliver physical megaprojects—where constraints are real, failure is expensive, and evidence matters.

Inspect the work Read the résumé
01 ↗upstreamOSS PRs mergedOpenClaw + Clawpatch · live query
026M lbrebar operation led$100M ship-lift drydock
03$1M+field savings deliveredthree months ahead of schedule
044frontier model familiesone production chat surface

Live product surfaces

Live right now.

Every tile is a real, operated surface — designed, built, deployed, and kept running. Click through to the live site.

Sunny Dayz storefront homepage
FEATURED / sunnydayz.coNext.js · Stripe · MongoDB

Sunny Dayz

A commerce storefront with its own operations dashboard: product import, margin rules, generated ad copy and photography, and Stripe checkout.

Open the live site

The build-and-operate rhythm behind these lives in theProduct Studio case study.

Operating schematics

The systems behind the products.

Three mechanisms I run in production, replayed as deterministic schematics: one iMessage number serving isolated tenants, a machine fleet sharing one memory, and parallel agents shipping without collisions.

SCHEMATIC / REAL MECHANISMS DETERMINISTIC REPLAY

Selected mechanism

One number, every tenant.

Group chats and DMs arrive on one iMessage identity. Every message must resolve to exactly one isolated tenant — or be rejected.

  1. 01inbound · dm · sender paired
  2. 02broker · route → tenant-a · cas ok
  3. 03inbound · group "demo crew" · claimed
  4. 04broker · route → tenant-b · owner verified
  5. 05inbound · sender unknown · no route
  6. 06broker · ambiguous → REJECT · fail closed
DM · COLEGROUP · DEMO CREWDM · NEW USERDM · UNKNOWNONE NUMBERimessage / smsBROKERcas route tablejournal · replayTENANT ATENANT Bgroup claimed ✓TENANT CREJECTEDfail closed

Selected systems

Evidence over adjectives.

Professional systems, operated products, and public code—described through the constraint, the system, and what changed after it shipped.

X-energy / advanced nuclearCurrent role

Enterprise AI Systems

Turned domain expertise into a software role by repeatedly shipping useful systems from inside the engineering workflow.

TypeScriptReactServicesAgentsGitLab CI
Open case study
Engineering automation / X-energyInternal system

Excalibur

Converted a specialist calculation workflow into a repeatable, testable application with streaming progress and packaged delivery.

ReactFastAPIPythonSSEACI 349
Open case study
Badland AI Chat model picker and composer in dark mode03 / SELECTED
Multi-model product / operated in productionLive

Badland AI Chat

A production product that lets the interaction choose the model workflow—not the other way around.

React 19Express 5MongoDBtRPCAI SDK
Open case study
Badland Agent Platform public iMessage product demo04 / SELECTED
Personal agents / native messagingPrivate beta

Badland Agent Platform

Moved the agent from a separate app into the communication surface people already use—then hardened the transport behind it.

imsgOpenClawTypeScriptTenant isolationSMS
Open case study
ai-cli multi-provider council in a terminal UI05 / SELECTED
Rust / multi-provider terminal UIActive tool

ai-cli

Makes model disagreement visible without turning provider routing into a pile of shell scripts.

RustRatatuiStreamingWeb searchLLM
Open case study
OpenClaw + Clawpatch / 2026Merged upstream

Upstream Contributions

Contributions focus on the failure paths that make agent tooling dependable for other operators.

OpenClawClawpatchTypeScriptPythonReliability
Open case study

Full project ledger

The useful work continues past the highlights.

01

Professional systems

  1. Enterprise AI SystemsEnterprise software and agentic workflows built where engineering rigor, security boundaries, and operational reliability are non-negotiable.Current role
  2. ExcaliburA full-stack reinforced-concrete design application for turning finite-element shell forces into reviewable ACI 349 and ACI 318 reinforcement designs.Internal system
  3. ExcitoA portable Python command-line workflow that modernizes legacy seismic-matching methods and connects them to contemporary analysis pipelines.Engineering toolkit
  4. Engineering AnalyticsA cross-program analytics pipeline connecting GitLab delivery data, project-controls schedules, and leadership reporting.Internal analytics
02

Independent platforms

  1. Badland AI ChatOne coherent chat surface for four frontier-model families, voice, images, files, web research, councils, and concise briefings.Live
  2. Badland Agent PlatformPersistent personal agents that work through native iMessage and SMS, with isolated tenants, memory, tools, reminders, and artifact creation.Private beta
  3. Badland Agent RuntimeThe runtime beneath the personal-agent product: tenant provisioning, tool boundaries, routing, observability, and recovery.Operated in production
03

Products & experiments

  1. ai-cliA Rust CLI and Ratatui interface that streams one prompt across multiple model providers for side-by-side comparison or focused single-provider work.Active tool
  2. Product StudioA continuing line of focused products and agent-built artifacts across communities, commerce, finance, personal dashboards, and small-business workflows.Ongoing
  3. Lux Voice AgentA Raspberry Pi voice interface with local wake-word detection and switchable real-time model backends.Prototype
  4. Airspace LabA dependency-light Three.js trainer for understanding terminal airspace by orbiting, slicing, probing, and testing the volume instead of memorizing a diagram.Public lab
04

Open source

  1. ai-keysA unified command-line interface for inspecting and managing provider API keys across machines.Prototype
  2. Upstream ContributionsPublic upstream pull requests merged across agent-channel reliability and resilient multi-provider review tooling.Merged upstream

Operating history

Software judgment, built in the field.

The through-line is delivery under constraints: first with concrete, crews, cost, and schedule; now with models, services, permissions, and state.

  1. 01

    X-energy

    AI Software Engineer

    X-energy tenure since Nov 2024
    Rockville, MD

    Building enterprise AI and engineering systems for advanced nuclear work. Moved into software after first shipping internal tools from within civil and structural engineering.

    • Build enterprise agentic workflows, platform capabilities, and full-stack engineering software with an emphasis on reliability, access control, and traceable results.
    • Designed and shipped Excalibur, a client/server reinforced-concrete design application that turns finite-element results into reviewable ACI 349 and ACI 318 reinforcement designs.
    • Built Excito, a portable seismic-analysis toolkit that modernizes legacy Fortran methods behind a Python command-line workflow.
    • Built cross-program engineering analytics and schedule-to-execution automations around GitLab, Power BI, and project-controls data.
  2. 02

    Badland

    Founder & Independent Developer

    2024 - Present
    Remote

    Designing, shipping, and operating multi-model software, personal agents, native messaging infrastructure, and focused products.

    • Operate a multi-tenant chat product spanning OpenAI, xAI, Google, and Anthropic model families, with voice, image, web, file, council, and briefing workflows.
    • Built a personal-agent platform delivered through native iMessage and SMS, with isolated tenants, persistent context, reminders, media, search, and artifact creation.
    • Replaced the retired webhook transport with a native imsg broker that fails closed on ambiguous routes, isolates attachments, and journals delivery for bounded replay.
    • Maintain the product, infrastructure, observability, deployment, and incident-recovery paths end to end.
  3. 03

    Kiewit Infrastructure

    Project Engineer

    Jun 2021 - Nov 2024
    Florida / South Carolina

    Led field execution, cost and schedule controls, quality, procurement, and constructability across heavy civil and industrial programs.

    • Managed installation of six million pounds of reinforcement on a $100M ship-lift drydock with a 20-person ironworker crew.
    • Finished the reinforcing operation three months early, reduced budgeted waste from 10% to under 1%, and delivered more than $1M in combined savings.
    • Worked across a bridge and roadway program, airport concrete repairs, and a $250M industrial EPC project.
    • Used cost-loaded Primavera P6 schedules and earned-value reporting to connect field quantities, cost, and schedule performance.
FOUNDATION
B.S. Civil Engineering · Clemson UniversityEIT · LEED Green Associate · Eagle Scout
2017—2021

Working set

Four connected disciplines.

The stack changes. The durable advantage is being able to cross product, runtime, infrastructure, and domain boundaries without losing the system.

01

Agent systems

Models become useful when identity, tools, memory, and recovery are designed together.

  • Multi-model routing
  • Tool boundaries
  • Native imsg
  • Evaluation
  • Observability
02

Product engineering

Own the user-facing path from first interaction through streamed work and durable state.

  • TypeScript
  • React
  • Node / Bun
  • Python
  • Rust
03

Production systems

Deployment is the start of the feedback loop, not the finish line.

  • Cloudflare
  • AWS
  • Containers
  • CI/CD
  • Tracing
04

Engineering domain

Physical constraints and reviewable calculations still shape the software I build.

  • ACI 349 / 318
  • FEA
  • Seismic analysis
  • P6
  • EVM

About

The unusual part isn’t the stack.

I learned engineering where mistakes become schedule slips, cost overruns, rework, and safety risk—not just failed builds.

Before moving into AI software, I managed field execution on infrastructure programs ranging from airport repairs and bridge construction to a $100M ship-lift drydock and a $250M industrial EPC project. I learned to read the whole load path: design intent, materials, people, sequence, money, inspection, and the recovery plan when reality disagrees with the drawing.

That background shapes how I build AI systems now. I specialize in agentic systems, multi-model tooling, and the infrastructure that makes frontier models useful outside a demo—authentication, isolation, observability, deployment, and reliability. At X-energy, that work sits inside advanced nuclear. Independently, I build and operate Badland and a continuing line of focused products.

My edge is not knowing one framework better than everyone else. It is moving between domain constraints and software architecture while keeping the result understandable to the people who have to trust it.

Start a conversationGitHub ↗LinkedIn ↗
01 / Find the real constraint02 / Make state inspectable03 / Fail inside the boundary04 / Verify after “done”