JEREMY PANASUK | THE CLOSER
Operator
Jeremy Panasuk
Available · AI Systems Engineer · AI Data Engineer

JEREMY
PANASUK

Informatica used to parachute me into Fortune-100 data platforms that were already on fire. I recovered 19 of 20 of those accounts. The last two years I rebuilt that career on purpose: off Informatica-only, onto a private RTX 4070 lab that measures decode, routes GPUs, and runs four isolated Python agents against local Ollama. The motion did not change. Hear the floor. Wrangle what exists. Ship a path the night operator can still run on Tuesday. Measure the number you named before you started.

Customer satisfaction is worthless. Customer loyalty is priceless. That is not a tagline. It is the talk I gave at Informatica while I was still embedded at Dawn Foods. Satisfaction is a survey they forget. Loyalty is the warehouse load that went from eight hours to under one and stayed there because they used it.

AI Systems Engineer · Enterprise Data Architect · 22 Years Fortune-100 · Private RTX Lab

Sand Creek, WI · 715-970-0204 · jpanasuk@gmail.com

27×Local decode after GPU routing: 1.39 tok/s CPU baseline to 37.47 on the RTX 4070, logged in SQLite. Not a lucky run.
8h → 1hDawn Foods warehouse load on PowerExchange CDC against AS/400 remote journals. Featured on informatica.com because it got used.
$2.3M+Bank of New York Mellon Grid/HA with senior VPs and Informatica CEO Sohaib Abbasi in the room. Sales, services, support, growing.
19/20Troubled accounts entered. Nineteen recovered. Milwaukee insurer: three-month pile-up closed in two days — one buried IBM DB2 driver.
Enter the floor IPS playbook Canonical stack packet
Trusted by PepsiCo · BNY Mellon · Mercedes-Benz · Farmers · Northwestern Mutual · Waddell & Reed · Toyota · Kohl’s · Navy Federal · Micron · Smith & Nephew · Ameriprise · JPMorgan Chase · Harvard · Cargill · Federal Reserve · IRS · US Navy · VA · DHS
Who I am

I build systems that stay alive under pressure. Twenty-two years of Fortune-100 integration work taught me to read the evidence before touching the config, find the bottleneck beneath the symptom, and leave the customer with a platform and a team that can run it. Consulting Manager: teams of 7–14, analyst through principal. About fifteen percent of IPS time was writing statements of work with regional managers — full contract wording, resource levels, estimates and assumptions — Micron 2,080 hours grown to 7,000, Navy Federal 3,000 hours, Harvard through hypercare. At JPMorgan Chase I sat sales, regional managers, and IPS VPs on enterprise data-warehouse and Center of Excellence roadmaps. We reached the final two proposals. We did not close. I will not write that as a $10 million win. I still parachute toward the fire.

The operating line

Customer satisfaction is worthless; customer loyalty is priceless.

Independent lab

AI Systems Engineer · Local AI Lab

A private RTX 4070 lab built with Docker Compose, CUDA, TabbyAPI/EXL3, Ollama, MCP, SQLite telemetry, and RAG. The point is not a demo. Every claim gets a benchmark, a regression check, and a documented operating path — the same loop I used on Grid/HA. Four agents run as four isolated Python processes, each with its own virtual environment, handing Pydantic JSON instead of free text: Systems names the constraints, Architect writes the contract, Synthesis generates against local Ollama, Validator checks schema and graph and self-heals until it passes. If one agent crashes, the other three keep running. n8n RAG pulls live competitive telemetry, writes a Discord briefing every run, and only fires WhatsApp when the score hits a threshold or Rank 1 territory. Use the selector below. The full-detail bullets are the receipts, not the chips.

Aug 2024 – Present
Canonical stack / packet

One lab. One path. Real receipts.

The canonical packet is the forward bridge from this campaign into the unified stack: 3 days to stand it up, 21 days to learn and teach it, one compose, one network, one GPU, and every UI on qwen through /v1. It carries the live packet registry, hunt status, the nine LinkedIn reference screenshots, and the distinction between projected learning work and verified receipts.

Open canonical stack packet
Day 1 · Chat path

qwen :1234

Open WebUI :3000 · SillyTavern :8000 · MCPO :8001 · SearXNG :8080. The prompt path is visible; qwen stays up.

Day 2 · Coding pack

Same /v1 contract

code-server :8443 · LibreChat :3080 · AnythingLLM :3002 · Lobe :3210. No second inference server to make a UI feel faster.

Day 3 · Memory / workflow

Operate and hand off

Qdrant · Chroma · n8n · Meilisearch · Mongo · Postgres. Then ship the packet, keep STATUS honest, and teach the stand-up.

The packet points back here as the campaign hub. Forward: open the packet · Back: inspect the lab · Back: inspect the production proof

Real-world operating path

The Playbook

The parachute left documents, not folklore. Every IPS engagement opened with an Environment Summary — owner and date on the cover, utilization first, what was already changed, what is still on fire. Close included a named skill-up and, where it mattered, a QA twin of production. The public artifacts: Northwestern Mutual Environment Summary, Mercedes-Benz Financial performance cheat sheet, Waddell & Reed Upgrade Roadmap v3, Farmers Phase 2 architecture review, PowerCenter 10.0.0 upgrade overview. Internal hostnames and credentials stay off this site. The same loop now runs on the RTX lab: measure, isolate, write the standard, coach the team. Volkswagen Group of America wanted the complete Grid/HA stack in two weeks — bonded dual NIC, Oracle RAC on InfiniBand, three 8-core PowerCenter servers. Live demo a day early. Four of their consultants shadowed and took notes. An unnamed customer would not buy PowerCenter and Data Quality from a vendor VM, so I installed it on their metal, smoked it, ran their use cases, and closed $375k the night before quarter end. That is the playbook. Select a surface below for the full-detail bullets.

IPS artifacts · 2017 floor
01 · Docs

Documentation standards

Owner and date on the cover. Utilization graphs before folklore. What was changed. What is still on fire. Grid/HA and the filesystem called by name — UNC is not high availability. The customer can execute the next change without you in the building. If the next engineer cannot run it, I wrote a diary, not a standard.

02 · Perf

Real-world performance

Baseline, then a named goal, then the bottleneck, then eliminate it, then repeat. Thread statistics before folklore. Do not partition a single-partition mystery. Cache large transformations in RAM so they stop spilling to disk — Mercedes-Benz Financial took 20–30% off job runtime that way and kept the cheat sheet. Isolation tests before you add another box.

03 · Coach

Customer coaching

Knowledge transfer is a deliverable, not a courtesy. Skill-up is named in the write-up — Farmers called Developer tool and Advanced XML. Whole stack plus GCS in one room so the next incident is not a mystery. The team is praised in the document they keep. Loyalty is the runbook they still run on Tuesday. Satisfaction is a survey they forget.

Open the full playbook

The receipts

The Legend Reel

The role, the awards, the rescue record, and the enterprise outcomes behind the headline — full-detail bullets from the long-form resumes, not dollar chips. Click any card or use the selector. Informatica SEAL: analyst to Consulting Manager, v7–v10 upgrade specialist, parachute closer. Twenty troubled accounts entered, nineteen recovered. Kohl’s is not a logo: Informatica v9 Grid/HA capacity and upgrade, target architecture on commodity (Cisco or IBM) and the cutover off AIX, current volumes plus a three-year growth curve, briefed through Senior VPs, about $2 million in software, proofs of concept on their data when the vendor VMs were not enough. New receipts from the IPS artifacts: Farmers Phase 2, Mercedes-Benz Financial 20–30%, Northwestern Mutual P1, Waddell & Reed v9→v10 roadmap. Volkswagen two-week complete Grid/HA. Micron 2,080→7,000 hours, zero downtime. The selector expands every named floor the way a hiring manager should read it: what was broken, what was built, who was briefed, what moved.

What they said
“Jeremy has been a pleasure and an asset to work with. His knowledge not only of the Informatica suite but also at the operating system level has been truly invaluable.”
Dawn Foods · Client
“He’s a rock star out at Dawn. The customer loves him and the sales team has been leaning on him hard.”
Informatica IPS · Sales
“It is the belief of the infrastructure staff that you have sent one of the best that Informatica has to offer in Jeremy.”
MF Global · Infrastructure Staff
“I trusted that Jeremy would deliver, with quality. He exceeded expectations at someone of his grade level.”
The operator range
AI · Python · Docker · CUDA · Ollama · TabbyAPI EXL3 · vLLM · llama.cpp · Hermes · MCP · RAG · SQLite telemetry · OpenClaw · Open WebUI
DATA · Informatica PowerCenter · Grid/HA · PowerExchange CDC · Talend 8 · Oracle · SQL Server · DB2 · Teradata · Linux · DR
SHIPPED · github.com/jpanasuk-netizen · huggingface.co/jpanasuk/tabby-tavern-stack · playbook.html · loyalty.html · wago.io/fsgbObGQJ
Now available

SEND THE CLOSER

Remote AI Systems Engineer · Agent Architect · Local LLM infrastructure. High-autonomy teams that want production discipline, not a demo reel. I will sit on the floor, write the standard, and leave a team that can run the next wave without me in the chair.

Open the operator floor
715-970-0204
jpanasuk@gmail.com · jpanasuk@yahoo.com
Resume · Hunt board · IPS playbook · GitHub · Hugging Face · Canonical Packet · Connect Your Life · LinkedIn
Sand Creek, WI · BBA MIS, University of Wisconsin–Eau Claire