The clip · Carex 6273 · Madison hybrid 3 days · Direct hire · Resume
Customer satisfaction is worthless.Customer loyalty is priceless.
This is not a slogan I put on a landing page. It is the title of a talk I gave at Informatica while I was still embedded on Dawn Foods’ floor in Jackson, Michigan. Carex job 6273 is hiring a founding Forward Deployed Engineer for a fintech startup partner — Madison, three days a week onsite, direct hire — to sit with customers, turn messy process into something that actually runs, and write the playbooks for a function that does not exist yet. I am applying as the person who has already done that motion for twenty years at Informatica Professional Services: embed, measure, leave a path the next hire can run without me on Slack.
You asked for communication, ownership, AI agents in daily work, judgment about when to configure versus build, and honesty on Epic Agent Factory. Factory is preferred. I have not shipped an agent inside Epic Chronicles. I will take it the way I took Grid/HA: on the customer’s floor, with a number before and after. Formal software engineering is not required on this posting. Delivery is. That is this résumé.
Hear the clip Open resume POCs → revenue The posting
The clickable PDF is the full packet. Page 3 is the clip. Pages 4–5 are the talk — Bank of America, Dawn, John, Trillium. Page 6 maps every line of job 6273. Page 7 is POC · SOW SUPERSTAR: Volkswagen two-week Grid/HA, the local PowerCenter / Data Quality install sales never does, and the statements of work (Micron 2,080→7,000 hours, Navy Federal 3,000 hours, Harvard through hypercare). Pages 8–11 are the IPS artifacts: Environment Summary, Upgrade Roadmap, Phase review, v10 test deck. Page 12 is the four-agent runtime on local Ollama, Epic honesty, and founding the function.
01 — How I embed
If they cannot run it Tuesday, I did not finish.
Every engagement I have run uses the same five steps. Hear what the operators actually do — including the unofficial definition of a “good run” and the night workaround somebody invented in 2011. Wrangle what already exists instead of waiting for a clean ontology. Ship something they can operate. Measure the number you named before you started. Leave them running it on Tuesday without you in the chair. That is founding Forward Deployed Engineering. That is also how I ran Informatica IPS embeds. A founding hire who only installs software is a contractor. A founding hire who leaves a path is the function.
The night operator is the customer
A workflow that “works in UAT” is customer satisfaction. The night operator still using it in month three is customer loyalty. If the customer’s ops team cannot run the agent — or the load, or the quote path — without me on the phone, I installed software. I did not finish the job this posting describes: partner, design, configure, build, deploy, support, and leave a reusable kit hire two can start on.
19 of 20 troubled accounts recovered
At Informatica I was the person they parachuted into accounts that were already on fire. Twenty troubled engagements. Nineteen recovered. One example: a Milwaukee insurer had a three-month production pile-up. We closed it in two days. The cause was one IBM DB2 driver setting buried in IBM documentation. Monitor, troubleshoot, improve reliability — you listed that as the job after go-live. That is the fire department. Founding FDE still does support. I still do it.
02 — Job 6273, answered in sentences
What you will do, and what I already did.
Your posting is specific. Here is what each ask maps to in work I have already done — not a keyword list, not a chip.
Partner with customers · uncover the real challenge
IPS embed. Dawn Foods, Navy Federal, Mercedes-Benz, West Bend Mutual. I sit with the people who run the process, not the slide about the process. Same facts, two rooms: a plant manager at 6 a.m. and Bank of New York Mellon with senior VPs and Informatica CEO Sohaib Abbasi in the afternoon. The unofficial definition of a good part is the requirement. The night workaround from 2011 is the requirement. I write those down on the cover of the Environment Summary so they stop living in Slack.
Design, configure, build, deploy, support
Workflows, schemas, integrations, lightweight apps, then hypercare. Configure first. Build only when configure will not hold. At Dawn I mentored Type 2 warehouse design — slowly changing dimensions, not a one-off mapping. At Navy Federal I wrote naming standards and service configuration so their team could operate the platform after I left. A schema is a contract. That is the same idea as Pydantic JSON between agents in the lab.
Define success before · measure after
You posted this as a requirement. It is already the first graph on an Environment Summary. Dawn: warehouse load eight hours to under one, featured on informatica.com because the number moved and they kept using it. Mercedes-Benz: longest workflows +200%, and TCS kept the path. Local lab: 1.39 tokens/sec on CPU baseline to 37.47 after GPU routing, logged in SQLite. If the number did not move, we shipped theater.
Recurring needs become reusable kits
You are founding a Forward Deployed Engineering function. Hire two has to start on page 4 of a playbook, not a Slack thread. TCS kept the Mercedes performance-tuning path. Farmers named Developer tool and Advanced XML in the write-up. Navy Federal got a technical-stack test strategy for go-live. That is “translate recurring needs into reusable playbooks,” not a slide about culture.
Feedback to product and engineering
At JPMorgan Chase I worked sales, regional managers, and IPS VPs on enterprise data-warehouse and Center of Excellence roadmaps — field notes back into product. If three customers hit the same wall, it is the next feature, not another hero week. We reached the final two proposals. We did not close. I will not write that as a $10 million win.
Epic Agent Factory — preferred, not claimed
Epic Agent Factory is a no-code builder inside Epic EHR, previewed around HIMSS 2026, wider later. Personas people mention: Art, Emmie, Penny. Chronicles remains source of truth. I have not deployed an agent inside Chronicles. I have deployed agents with boundaries, customer workflows with unofficial definitions, and go-lives in regulated rooms (Navy Federal, Smith & Nephew FDA-validated platform, Prime Medicare). I will take Agent Factory the way I took Grid/HA: embed, measure, write the standard, leave a team that can shape the next one.
03 — POC · SOW SUPERSTAR
Customers buy what they just watched run on their own servers.
A proof of concept is supposed to be a demo in a lab virtual machine. Customers often will not buy from a lab. They buy what they just watched run on their metal. Both of the stories below closed immediately after. That is how a statement of work gets signed, not a slide deck. Page 7 of the PDF walks them in full.
Volkswagen Group of America — complete Grid/HA in two weeks
They did not want a subset. They wanted the complete high-availability stack: bonded network with dual NICs, Oracle RAC sitting on InfiniBand, and three 8-core PowerCenter servers. The window was two weeks. I walked a live demo and finished a day early. Four of their consultants shadowed the entire install, taking notes, so they could run the next one. On the other side of the building a parallel Data Quality implementation was burning down. That was the pressure in the room — it was not my DQ miss. I do not put a dollar figure on this copy because we never logged one in the kit. The close was: they watched it run on their RAC.
Unnamed customer — PowerCenter and Data Quality on their machines
Informatica sales does not install PowerCenter and Data Quality locally on a customer’s systems. The official path is the VM lab. This customer would not buy from a VM. They wanted it on their metal. I installed it anyway, because a sales representative needed to clear quarterly quota and a two-week POC was the only thing that would get a yes. Full smoke testing. Use-case analysis against their data. It passed with flying colors. The deal closed immediately after. That POC pushed someone over their target for the quarter. The client stays unnamed here. The motion is the point: policy said no, the customer wanted yes, I found the path.
The SOW is the product after the POC
About fifteen percent of my Informatica time was writing statements of work with regional managers — not a one-line hour chip. Full contract wording. Resource levels from analyst through principal. Estimates and assumptions. Range was mom-and-pop duration all the way to Micron: a resourcing plan that grew from 2,080 hours to 7,000 across PowerExchange, PowerCenter, Fast Clone, and Big Data, delivered on time with extensions. Navy Federal: 3,000-hour V1 with Informatica RMO plus their HR. Harvard Version 9: resource plan through post-production hypercare, not just go-live. Toyota: scoped and resourced a multi-product engagement with regional managers and sales. The rule I held: understand the estimates, then lead the team to deliver inside them. A two-week POC that actually runs is how that SOW gets a signature.
04 — Four-agent runtime on local Ollama
You asked for AI agents in day-to-day work. This is not a chatbot screenshot.
On my lab machine I run four agents as four separate Python processes, each with its own virtual environment and tool loop, orchestrated against local Ollama on an RTX 4070 under WSL2. If one agent crashes or writes a bad script, the other three keep running. They do not chat in free text. They hand off structured JSON contracts. Page 12 of the PDF walks the architecture. This is the same self-heal loop I use when I bench GPUs in Local Grid Suite.
01 Systems
Names the constraints and interfaces: what already exists, what must not break, what “done” means. Same job as the cover of an Environment Summary — owner, date, utilization first, what is still on fire.
02 Architect
Turns that into a Pydantic contract — a JSON schema the next agent can consume without guessing. No prose handshake. A schema is a contract on a bank floor and on a GPU.
03 Synthesis
Generates the payload against local Ollama. The work happens on this machine, the same way a POC happens on the customer’s metal instead of a vendor VM they will never buy from.
04 Validator
Checks schema and graph. If it fails, it writes, executes, and debugs Python through the local LLM until the contract passes. Isolation plus a self-heal loop. Not “I played with agents.”
05 — The clip, in full sentences
Satisfaction is the lowest acceptable service. Loyalty is Tuesday.
Gitomer. Carnegie. Greene. I gave this talk while still on Dawn’s floor. Two complete stories live in the PDF. Here is what they mean for a founding FDE in Madison.
Bank of America — they chose policy
I had been a Bank of America customer since 1999. Lost card. Ben on the phone could not click a button. Kari cited policy and could not help. Two clicks would have kept a lifetime of deposits, a home loan, a car loan. They chose policy. I left, and I took the word of mouth with me. Your posting says the customer wants measurable value and a partner who finds a path. Policy that leaves them with “the model can’t” is Kari. Founding FDE: never leave the night operator with Kari.
Dawn Foods — they still quote the number
At Dawn the warehouse load used to take eight hours. PowerExchange on AS/400 remote journals took it under one. They put it on informatica.com because it got used, not because we surveyed them. The BI lead, John, had four daughters. I knew that because I asked. Over lunch he told Trillium he did not even want a proposal — he would wait for Informatica data quality. That is not a close. That is a relationship that closed the next product by itself. Loyalty is Tuesday. CSAT is a survey they forget.
06 — Kohl’s, not a dollar chip
Capacity, upgrade, AIX off, Senior VPs, then the close.
When I say Kohl’s I do not mean a logo on a trusted-by line. I built the future data-platform utilization roadmap: Informatica v9 capacity planning and upgrade, including multiple high-availability and Grid components. I designed the target Grid/HA architecture on commodity hardware — Cisco or IBM — and the cutover off AIX. I analyzed current volumes and projected three-year growth, and sized the platform to that curve. I delivered the roadmap through multiple org levels up through Senior VPs. That work closed about $2 million in Informatica software. When the vendor VM labs were not enough, I ran proofs of concept on their own data for Data Quality and MDM. That is the kit. That is also how a founding FDE should talk about a named floor: what was missing, what was built, who was briefed, what moved.