You built this fellowship for the American worker who never left the floor. I never left. Now I want Foundry and AIP in my hands.
Trusted on the floor Cargill · Horizon Milling · Dawn Foods · Micron · Mercedes · Toyota · PepsiCo · Ace · Kohl’s · Navy Federal · Smith & Nephew · BNY · JPMorgan
You asked for builders who will wire a sensor, flash firmware, or ship production code. Who read a messy CSV and think instrument, automate, optimize. Who do not wait for perfect specs. Twenty years on live plants, then two years of Docker, WSL, Python, and agents until the numbers showed up in SQLite. This fellowship is 12 weeks of Foundry and AIP. I will treat it like a deployment.
Cargill oils plants: BA overseeing plant ERP implementation — live operations, not a sandbox. Horizon Milling: production visibility on a time dimension so plants and accounting saw the same facts — CFO Award 2005 (six awarded that year). Dawn Foods PowerExchange CDC on AS/400 remote journals: warehouse load 8h → 1h, Type 2 DW, featured on informatica.com. Micron: connected fabs with PWX CDC off SQL Server, Fast Clone, BDE into Teradata/Hadoop. Mercedes +200% on the longest workflows; playbook stayed with the team. Toyota Engine Management go-live, all consultants 5/5.
Secret Key Class B shooter, 1996–2004. Fireworks Forever inventory manager and sales lead — $800K+ seasonal stock, 6 consecutive years of profit growth, 100-hour peak weeks. Load, wire, shoot, tear down. No schema, no second take. Then I put a private RTX 4070 in the garage and measured it. You asked for garage-shop builders who automate a physical process. I already lived the shop. Then I instrumented the stack.
Daily. Tabby Tavern on WSL2 + CUDA: TabbyAPI EXL3, Ollama, Open WebUI, SillyTavern, SearXNG, flash-attention, q8 KV cache. local_grid_suite: discovery, decode/prefill benches, GPU routing, SQLite, CLI + PDF. Headline 1.39 → 37.47 tok/s after routing (~27×). Warm stream ~76 tok/s. Dockroot-MCP. OpenClaw WhatsApp agents, 10+ skills. TypeScript for this packet. I taught myself the new stack the same way I learned the plants — by shipping.
AS/400 journals the MES does not see. SOAP that never matched the WSDL. Unofficial definition of a good part. Demantra demand planning — won Quarterly Award for Excellence, Oct 2006, separate from the oils ERP. Milwaukee insurer: platform dark three months; I closed it in two days on one buried IBM DB2 driver. 19 of 20 crisis accounts recovered. Prototype, test, learn on the fly. Perfect specs never arrived. That is the fellowship muscle.
Containerized private inference on RTX 4070 / WSL2: Docker Compose, CUDA, GPU passthrough. Bring-up, auth, and security notes treated as ops, not a demo folder. huggingface.co/jpanasuk/tabby-tavern-stack
Operator telemetry: model discovery, decode/prefill, GPU routing, SQLite logs, CLI + PDF reports. Constant test loops — not one lucky run. github.com/jpanasuk-netizen/local_grid_suite
Bounded agent visibility into Docker services and network wiring — not host-root. Same production discipline as Fortune-100 Grid/HA: measure the boundary, then hand the agent the boundary.
Systems → architect → validator over local Ollama: structured output, repair retries, collision checks. Production WhatsApp agents. Multi-provider routing with cloud failover. Hermes Agent on the same box.
Also retained through 2025 for Informatica production rescue: Grid/HA, PowerExchange CDC, IDQ, v9/v10 upgrades, SOAP/REST tuning, SOW/sizing, install-to-go-live. Operator floor: jpanasuk-netizen.github.io/the-closer
I do not ask for a clean schema. I walk the line, find the unofficial definition of a good part, and pull the journal the MES does not see. What broke last Tuesday at 2 a.m. is the ontology. Palantir’s job is to make that ontology executable on Foundry and AIP. That is the work I already do — I want your stack, not a toy dashboard.
I will not fake Verilog or ladder logic. I will wire, flash, and learn whatever the plant needs. Palantir asked for hunger, not a fake FPGA resume. I have Linux, Git, containers, production Python, SQL, and twenty years of physical operations. The rest I will take from Ontologize and the floor.
Talend 8 mappings from simple through complex on a live production platform. GitLab CI/CD migrations. Amazon Redshift + SQL Server tuning. Sustained 99.99% stability. On-call incident command and root-cause writeups. Messy production, not a sandbox — the same motion the fellowship wants on Foundry.
PepsiCo — Grid/HA + Data Integration Hub across NA/LATAM. Led v9→v10 with zero data loss. Teradata → Big Data (~40% analytics lift). Follow-the-sun with offshore. Ace Hardware — Teradata warehouse → Big Data objects and runtimes. Business + QA deploy path; sat with the floor, not just the ticket queue. Ameriprise — Enterprise Data Governance: IDQ, EDC, Axon. 200+ critical data elements. 24×7 platform, patches, local DG rollout.
Started as Analyst in 2008. Analyst → Sr. Analyst twice in four years. Sr. Analyst → Program Manager in ten. SEAL team. Teams of 7–14. v7–v10 upgrade specialist. Consultant of the Year runner-up (2016). Parachute closer: 20 troubled accounts entered, 19 recovered. The original forward-deployed motion — different badge. I sat with manufacturers, banks, and fabs from the line to the exec table and left a system the next shift could run. Ladder: Sr. Analyst → Principal Consultant → Technical Delivery Manager → Solution Delivery Manager → Program Manager / Consulting Manager.
Technical lead for Medicare and Medication Therapy Management data marts. Analyst → Sr. Analyst on a 4/5 annual review; internal 5-star on trust. Delivered above grade-level. Quality first — the quote below is theirs.
Oils plants: business analyst overseeing plant ERP implementation — live mill operations, not a PowerPoint ERP. I do not remember the product name; I remember the floor. Demantra (separate): demand planning. Won the Quarterly Award for Excellence, Oct 2006, for that work. Horizon Milling: production visibility on a time dimension so plants and accounting could see the same facts — CFO Award 2005. 24/7 warehouse / ODS support. This is the American manufacturing data Palantir wants on an ontology.
Professional display work: load, wire, shoot, tear down. $800K+ seasonal inventory. Six consecutive years of profit growth. 100-hour peak weeks. No schema, no second take. The American workshop Palantir is writing hymns about — I already lived it. Customer loyalty is priceless. High-autonomy only.
Code · Python · SQL · JavaScript/TypeScript · Docker · Linux / WSL2 · Git · GitLab CI/CD · Bash · APIs
Data / AI · Informatica PowerCenter · Grid/HA · PowerExchange CDC · IDQ · EDC · Axon · Talend 8 · Teradata · Hadoop · Redshift · Oracle · SQL Server · DB2 · Ollama · TabbyAPI EXL3 · CUDA · MCP · RAG · OpenClaw · SQLite telemetry
Physical ops · oils plants · plant ERP implementation · Horizon Milling · Demantra demand planning · food · auto · fab · CPG · Class B fireworks · $800K+ inventory
Education / certs · BBA, Management Information Systems — University of Wisconsin–Eau Claire · Informatica Certified Specialist: Administrator · Big Data Administrator · PowerCenter Developer I & II
This posting · 2+ years (20+) · Python · Linux shell · Git · Docker · WSL · no degree required (have the BBA) · remote · unpaid 12-week cohort · AIP / Foundry via Ontologize · U.S. person · available Fall 2026