The episode is a wide-ranging conversation with Franz MJ Pfister, CEO of DeepC France, starting with light chat about travel and the RSNA conference (and how chaotic it can be with weather, cancelled flights, and the “booth size = momentum” optics). From there, the host frames the real theme: radiology AI has moved from “cool algorithms” to the harder problem of making AI work reliably at scale inside real hospital workflows.
Franz explains DeepC’s origins: he’s a medical doctor by training with an AI background, and the company initially experimented with anomaly-detection models in collaboration with researchers. When they tried to operationalize those models clinically, they ran into the messy reality of deployment—local infrastructure, PACS connectivity, servers/GPUs, security, and instability. That pain revealed a bigger gap in the market: not another point-solution model, but infrastructure that makes many models deployable, governable, and usable in day-to-day practice.
DeepC’s strategy evolved accordingly: what began as a distribution/procurement “app-store” style layer (access to many algorithms through one integration) wasn’t enough, because access alone doesn’t improve outcomes or productivity. Franz emphasizes the next phase is orchestration—running the right AI across workflows, vendors, geographies, and regulations; measuring impact over time; and avoiding the N×N integration explosion that hospitals face when they try to manage many tools, contracts, support paths, and hosting setups. He compares this shift to earlier eras of cloud/data/devops: once something becomes mission-critical, infrastructure matters more than features.
They connect this to market forces accelerating the change, especially foundation models. Franz argues foundation models increase competitive pressure (new entrants can move faster), decentralize development (researchers can build narrow models without forming venture-backed startups or pursuing full regulatory commercialization), and invite generalist, well-funded players into healthcare AI. The net result is more models, more fragmentation, and more need for platforms that absorb complexity rather than pushing it onto clinicians and IT teams.
A concrete example is DeepC’s “ADA,” which Franz describes as a desktop “shell” that becomes powerful once connected to data sources, their gateway, LLMs/agents, curated medical knowledge, and potentially EHR context. In practice, it can notify a clinician about urgent findings, jump them into the images, pull relevant history/labs/meds, reference trusted clinical knowledge, and help draft structured reporting and next-step recommendations—turning AI from reactive detection into assistive, context-aware workflow support. The conversation closes with Franz’s founder lessons on scaling carefully (culture and ownership break when hiring too fast), hiring for “stage fit” over pedigree, and advice to aspiring founders: follow what you’re genuinely passionate about, build in that direction, and the path becomes clearer through real work and connections.
