ORVIS MISSION CONTROL // PUBLIC DASHBOARD
Dalton
Orvis
FORWARD DEPLOYED AI ENGINEER
I put AI agents into production and keep them there. At IdeaRoom I was the only engineer behind Dorvis, an agent platform about 20 teammates flew daily across CX, implementation and sales. It worked about 1,500 customer tickets in six months. Now I'm looking for my next mission.
Flight-Proven Numbers
~0
TICKETS WORKED
Customer tickets worked by Dorvis in six months
~0x
FASTER
Agent runs vs. the human estimate, first six weeks
~0%
LOWER COST
vs. the human estimate for the same work, first six weeks
~0
DAILY OPERATORS
Teammates across CX, Implementation, ProServ and Sales handoff
SOURCE: OPS-1 DORVIS AGENT PLATFORM · IDEAROOM-3 MISSION DATA · MAR TO OCT 2026
Missions Flown
NEXT-4
CREW NEEDED
AWAITING LAUNCHMISSION BRIEF
Looking for my next mission
Forward deployed, founding AI engineer, or agentic systems roles. I do my best work embedded with the people who will use what I build: finding the work worth automating, building the agent, and running it in production until the team trusts it.
ESTABLISH UPLINKIDEAROOM-3
DURATION 0Y 10M
HANDED OVERIdeaRoom, Inc.
JAN 2026 / OCT 2026
DIRECTOR OF AGENTIC SYSTEMS, EXECUTIVE LEADERSHIP TEAM
Came in as a consultant in January to test whether AI could clear a growing support backlog, and joined the executive team full time in March on the results. As the sole engineer of the agentic systems function, I took that pilot to Dorvis, a production agent platform, and handed it over in October with infrastructure as code and 280+ docs and runbooks.
- Built and owned the Dorvis agent harness. Replaced a heavily patched open-source fork (about 42,700 lines of local changes) with a lean TypeScript engine on the Vercel AI SDK, with a sandbox per conversation and a Langfuse trace on every run. A replay suite of about 50 real cases proved it first: 2x faster on 65% less compute, at a quarter to a third of the cost.
- Context engineering: raised cache reads from 32% to about 90% of input tokens, cut payload tokens 30 to 40%, swapped per-turn memory injection for on-demand recall, and built an agent-facing CLI over IdeaRoom's APIs. Token cost per ticket fell about 55%.
- Built Dorvis Triage, a headless agent that picks up Intercom tickets and proposes an action plan for an operator to approve. Tested it in a simulation harness before real users, launched a week early, and v2 reached 11 of 12 teammates within 10 days.
- Moved the platform from a single machine to AWS (ECS Fargate, RDS Postgres, EFS, S3, Cognito SSO) in about 5 working days, with blue/green deploys, automatic rollback and PR review environments. Ran on-call and wrote the RCAs.
- About 1,500 customer tickets and 7,200 catalog edits in six months, an estimated 1,200 hours of ticket work. Helped cut the support backlog from about 173 to 78 open tickets.
- Wrote about 200 architecture decision records, trained three departments, led the session that produced Engineering's agentic-first best practices, and shipped 416 merged PRs in five months.
TypeScriptVercel AI SDKLangGraphLangfuseVercel SandboxAWS ECS FargateRDS PostgresSSTClaude APIOpenAI APIRQ-2
DURATION 0Y 6M
COMPLETEReliaQuest
SEP 2025 / MAR 2026
SOFTWARE ENGINEER
Short, high-tempo rotation on a cybersecurity platform. The mission objective shifted from features to fleet enablement: getting AI into the team’s daily workflow.
- Integrated AI-powered code reviews into the CI/CD pipeline, catching issues earlier in the cycle
- Ran the weekly AI training session teaching senior and mid-level engineers AI-driven workflows
- Shipped AI agent systems into production workflows for automated analysis and decision support
TypeScriptReactJavaPythonGraphQLLLM APIsVERISK-1
DURATION 2Y 7M
COMPLETEVerisk Analytics
FEB 2023 / SEP 2025
SOFTWARE DEVELOPMENT ENGINEER IN TEST II
First long-duration mission. Co-architected a department-wide QA automation platform and led the org’s first serious evaluation of AI tooling.
- Co-architected a department-wide automation app (Next.js + ASP.NET) saving 30+ minutes daily per user across 20+ engineers
- Built the first GraphQL contract testing architecture in the flagship Xactimate Online product
- Architected a C#/.NET API client library (NuGet) that moved the department from UI-first to API-first testing
- Led the QA org’s AI tool evaluation: research and proofs of concept that guided executive adoption decisions
- Promoted to QA Team Lead for backend services within the first year
C#/.NETNext.jsGraphQLContract TestingAI Tool Evaluation
- 2020 / 2023
- SCINTILLANT BIOSCIENCE
- Program origin. Biology labs, automating data workflows with Python/Pandas: 30-minute manual processes down to sub-second runs.
- 2020 / PRESENT
- INDEPENDENT OPERATIONS
- 5+ production applications shipped across e-commerce, SaaS, and enterprise, including NutriSimple. RAG on Pinecone, multi-tenant platforms, Stripe and PayPal payments.
- 2023 / 2024
- NINE DESIGN + WEB
- Enterprise social platform on Laravel. Redis queues cut response times 65%; multi-tenant isolation for 20+ enterprise clients.
- 2022 / 2023
- MAIDEN VOYAGE SOFTWARE
- Laravel/Vue.js CRM handling 10K+ daily interactions: 3x faster page loads, 80% less database load.
PROGRAM ORIGIN: biology labs, automating data analysis with Python. That foundation in complex-systems thinking still steers the program: understanding when emergent behavior adds value, and when it creates unpredictable risk.
Flight Rules
Mission control keeps flight rules: decisions written down before the pressure hits, so you don’t improvise them at 2 a.m. These are mine, learned across enough launches to know which failures repeat.
- FR-01
Sit with the people doing the work before you build anything.
The best agent use cases came from watching a support engineer spend an hour on one ticket, not from a roadmap.
- FR-02
Measure before you ship.
A golden dataset and a replay suite told me the new harness was faster and cheaper before anyone used it.
- FR-03
Build the error handling and oversight layer before the flashy parts.
Production AI that fails silently is worse than no AI.
- FR-04
Autonomy is earned.
Agents propose and a human approves until the evals and the operators say they can do more.
- FR-05
The agent isn’t done when it works. It’s done when the team trusts it.
Onboarding is part of the deliverable, including knowing when not to use it.
- FR-06
Write the decision down.
About 200 decision records at IdeaRoom, so the next engineer knows why, not just what.
Establish Uplink
Hiring for a forward deployed or founding AI engineer? Building agents and want to compare notes? Open a channel. I read every transmission.
Looking for my next mission. I work best embedded with the people who will use what I build: finding the work worth automating, building the agent, measuring it against real cases, and running it in production until the team trusts it.
EXPECTED RESPONSE: WITHIN ONE ORBIT (24–48H)
AUXILIARY CHANNELS