Wojtek
Grabski.

At Aymes Medical (1995–2004), my work combined medical reporting software, IT management, installation, and technical, customer, and sales support. At Hess Packaging (1995–1996), I worked on business databases and production management.

Working close to users taught me to explain technical choices in terms of the job at hand.

Databases / Reporting / Deployment / User support

At the Affect and Cognition Lab, I worked on research computing, imaging, and experimental systems. I also worked on the mathematics of early machine-learning with emotional facial-expression recognition, including work with students from Geoffrey Hinton’s lab.

My work was in cognitive neuroscience. I sure did miss the early boat on ML, but I gained an intuitive understanding of the mathematical foundations and cognitive processes that inform how I approach machine learning today.

Scientific computing / Cognitive systems / Imaging / Applied mathematics

I founded Flux Inc. around practical problems in medical imaging. DICOM Printer was being sold and integrated into imaging workflows in 2007; development, sales, implementation, and support informed one another from the start.

I spend much of my time coding and designing systems around people. Product direction, commercialization, and management sit alongside that hands-on development work.

Over the years, that work grew into products serving thousands of facilities and relationships with large instrument vendors and hospital networks.

My responsibilities span hiring and team building, sales and partner communication, marketing, contract negotiation, and commercial and regulatory work. I translate between what a customer needs, what a team can deliver, and what the business can sustain.

Hands-on development / Company leadership / Commercialization / Team building

For Yellow Pages, I built and managed high-volume website and media-production workflows: intake, copy, design, staffing, quality checks, and delivery across multi-discipinary teams.

I worked on change control, quality assurance, capacity planning, and staff training as the production mix changed.

These systems depend on clear handoffs: what information is missing, who owns a decision, and when work is ready to move on.

Production leadership / Capacity planning / Staffing / Marketing & copy

I co-founded Influx and built the product from the ground up. I designed the clinical software and the human workflows around it, connecting patient intake and scheduling to clinical reporting, billing, and financial reconciliation.

That meant understanding how clinical staff, administrators, and finance used the same information—and designing the handoffs between those roles.

I also designed the marketing and branding, managed trade-show attendance and sales channels, and took part in board-level work. Product design, implementation, and commercialization were parts of the same job.

Product work / Co-founder / Workflow architecture / Commercial leadership

I built reporting software for radiology, applications for content and commerce, and integration designs for pharmacy workflows. Information presentation mattered as much as data movement: each specialist needed the right view and the right next step.

For DXA Desk, rollout planning covered staff training, test exams, a paper fallback, and clinical acceptance. A pharmacy and genetics integration design for GenYouIne addressed specialist roles and staffing as well as interfaces. At Informedy, I owned the product from proposal through implementation and handoff.

Good software has to fit the job. That includes what happens before someone opens it and after someone closes it.

Product ownership / Specialist interfaces / Integration / Rollout planning

Medical imaging has made reliability a central part of my career: continuous processes, distributed image delivery, reconciliation, redundancy, and recovery when a component fails.

I created DICOM Capacitor and contributed to DICOM Printer 2. Alongside the code, I worked on specialist workflows, information presentation, customer integration, and support across secure environments and hospital networks.

A running process is not a delivered image. A delivered image is not a completed clinical workflow. I follow the chain through to the person who depends on it.

Distributed systems / Image delivery / Fault tolerance / Clinical interfaces

I began connecting AI to clinical tools and company operations. A Python DICOM MCP prototype exposed node discovery and connectivity checks to an AI assistant; a separate workflow used AI to reconcile fragmented purchase and installation records.

I designed a model-governance layer around role-aware context, memory, authority, and company rules. With the Flux Inc. team, that direction developed into retrieval, conversational rules, and compliance mechanisms.

The product question was broader than model access: what does an assistant need to know about the organization, what can it do on someone’s behalf, and how can people direct and correct it?

Product direction / Clinical integration / Context architecture / Business automation

I develop the tools and operating practices for AI adoption at Flux Inc.: shared company knowledge, role-aware workspaces, staff training, onboarding, and supervised support drafting. I connect infrastructure decisions with how staff will use, review, and maintain the system.

An investment trading desk was one predecessor to this work. Coordinating specialist agents, maintaining shared state, and recovering interrupted work in high-risk contexts helped give rise to Clawdapus and Cllama, for orchestration, observability, shared governance and cost control.

Tools for context and collaboration

Work that outlives a session

Flux Orchestrator Development pilots

Durable work records for people and agents: assignments, handoffs, questions, human review, and recovery after interruption. The task survives when a session ends.

Flux Gate Controlled access pilot

Human-approved, time-limited access to operational systems, with expiry, revocation, and an audit trail. Delegated work needs explicit authority as well as useful tools.

These tools combine original design and AI-assisted implementation.