Client delivery with a virtual technical team. Quality, timeline, unit economics, and a status the client can trust.
Delivered with Dawn · West Bend · Navy Federal · Mercedes · Toyota · Micron · PepsiCo · Ace · Kohl’s · BNY · JPMorgan · Farmers · NWM · Cargill · Harvard · Smith & Nephew
Toloka’s AI Data Project Manager owns client delivery: quality, dates, unit economics, and a virtual team of engineers and supply. That is Informatica IPS / SEAL — consulting delivery at scale, client in the room, experts and operators on a follow-the-sun clock. Last 24 months I ran the same loop on LLMs I actually measured. Remote US. Available. I do not have a PhD. I have the delivery seat.
Quality, timeline, cost, transparency. Dawn 8h→1h because it got used. PepsiCo v9→v10 zero data loss. Micron $3M extensions, zero downtime. Status the stakeholder can defend.
Teams of 7–14. Engineers, QA, GCS, offshore, the floor. Follow-the-sun. Named skill-up. Retros were writeups the next morning, not a slide.
Consulting background — the plus you listed. Clients asked for me by name. “Rock star.” Embedded until adoption, then hypercare.
Environment Summary. Unofficial definitions. Scope, risks, unit of work. I do not wait for a clean spec. I write the path the team can run.
Implementation, quality gates, throughput. Six-class test. QA twin of production. If humans do not run it Tuesday, it did not launch.
Named skill-up. Process change as a deliverable. TCS kept Mercedes docs. Playbooks are how 19/20 stays 19/20.
TabbyAPI EXL3, Ollama, Open WebUI, Compose + CUDA on RTX 4070 / WSL2. Same instinct as a data pipeline: measure decode before claiming quality.
Discover, bench, GPU routing, SQLite, CLI + PDF. 1.39 → 37.47 tok/s after routing. Throughput you can report — not a vibe.
Systems → architect → validator on local Ollama: structured output, repair retries. Production WhatsApp agents, 10+ skills. Multi-provider routing with failover.
Dockroot-MCP: agents see Docker/network, not host-root. Safety as a boundary, not a slogan — the same muscle as a red-team gate: what the model is allowed to touch.
Honest on SFT / eval / red-team: I have not run a labeling marketplace producing thousands of prompts a week. I have owned data quality, throughput, and client delivery on live platforms, and I have spent 24 months on LLM generate / evaluate / bound loops I can show. I switch domains the way IPS switched clients.
20 troubled accounts. 19 recovered. Milwaukee insurer: three-month outage closed in two days. Mercedes +200% on the longest workflows; playbook stayed. That is risk highlighted and resolved before the client stops trusting the weekly. Toloka’s delivery seat is that motion with Experts + Crowd + Tech.
Talend 8 on a live platform. GitLab CI/CD. Redshift + SQL Server. On-call incident command and root-cause writeups the next morning. Throughput and quality on a clock — the same operating loop as an eval pipeline with a weekly SLA.
PepsiCo — Grid/HA + DIH across NA/LATAM. Led v9→v10 with zero data loss. Teradata → Big Data (~40% analytics lift). Follow-the-sun with offshore. Tuning written into go-live timelines. Ace Hardware — PowerCenter → Big Data with business + QA. Ameriprise — EDG: IDQ, EDC, Axon. 200+ critical data elements. 24×7. Data quality as a control process, not a hope.
The consulting plus you asked for. Analyst → Program Manager in ten years. SEAL team. Teams of 7–14 engineers and specialists. $4M+ services in two years. Client communication + internal coordination as the default. Discovery, decomposition, scope, implementation, adoption, hypercare. Consultant of the Year runner-up (2016). Parachute closer: 20 in, 19 recovered. Ladder: Sr. Analyst → Principal → TDM → SDM → Program Manager / Consulting Manager.
Also: Merrill / Time Inc. SFDC lift-shift. Ivy Funds Grid/HA. Fed Reserve Philadelphia. Giesecke REST/SOAP, VMs, GitHub, Docker. Ameriprise 200+ CDEs. Retained through 2025 for production rescue: Grid/HA, CDC, IDQ, v9/v10, SOAP/REST, install-to-go-live.
Medicare + MTM data marts. EDW test lead. Analyst → Sr. Analyst on a 4/5. Quality first — the quote below is theirs. New regulated domain, shipped.
Plant ERP on live operations. Demantra: Quarterly Excellence Oct 2006. Horizon Milling: CFO Award 2005. 24/7 across 20+ warehouses. Domain switch is not a theory.
$800K+ seasonal inventory. Four full-time sales. Profit up six years. Crowd + operators + a clock. Loyalty is the metric. Toloka’s Experts + Crowd + Tech is that motion with models on the other side.
Delivery · client-facing consulting · virtual XF teams 7–14 · discovery / decompose / scope / implement / adopt · retros · status · unit economics
Data quality · IDQ · EDC · Axon · DVO · 99.99% platforms · zero-data-loss cutovers · QA twin of production · incident command
LLM / AI · local generate–evaluate–bound loop · TabbyAPI · Ollama · RAG · structured agents · SQLite telemetry · MCP boundaries
Code / data · SQL · Python · Informatica PowerCenter · Grid/HA · PowerExchange CDC · Talend 8 · SOAP/REST · Docker · GitLab CI/CD
This posting · U.S. remote · BBA MIS, UW–Eau Claire, 2004 · Informatica Certified Specialist (Admin, Big Data Admin, PowerCenter Developer I & II) · no PhD
Honest · Have not run a labeling marketplace SFT mill. Have owned client delivery of data systems at scale, and 24 months of measured LLM work I can show.
Operating line · Customer satisfaction is worthless. Customer loyalty is priceless.