Fika Jobs, a young Stockholm-based startup, just closed a $4 million round to build a recruiting platform that feels like a cross between LinkedIn and TikTok. Its core offering consists of short video profiles of candidates and AI agents that actually conduct job interviews, partially replacing human recruiters.

The idea is far from trivial. Video promises to restore the expressive dimension that a plain-text CV flattens, while conversational agents powered by large language models – LLMs – can ask questions, assess answers and deliver an initial screening in an ostensibly neutral way. The result is a faster hiring pipeline and, ideally, fewer biases.

The data tangle nobody talks about

Yet there is a detail that anyone managing recruitment in regulated sectors – banking, healthcare, public administration – cannot ignore. An interview handled by an AI agent produces, processes and stores personal and often sensitive data: from voice to facial imagery, including automated evaluations. When the infrastructure running these agents is a standard public cloud, effective control over data processing becomes complicated.

This is not a theoretical concern. The European GDPR imposes strict requirements on data residency and automated decision-making transparency. A candidate has the right to know if they were rejected by an algorithm and to challenge its logic. When the entire process relies on third-party APIs and hosted models, the chain of accountability lengthens and auditing turns into an obstacle course.

The on-premise alternative and its trade-offs

For an organization that wants to adopt similar tools without surrendering data sovereignty, on-premise deployment – running models on internal machines, possibly in air-gapped mode – is a viable path, though not frictionless. It requires adequate hardware, managing local inference pipelines, and coming to terms with the total cost of ownership (TCO), which includes GPUs, maintenance and in-house skills.

This is where the discussion broadens. Fika Jobs’ platform, however cloud-native by design, signals a larger movement: conversational agents entering sensitive decision-making processes. It is not science fiction to imagine large enterprises demanding self-hosted versions of these tools, especially in Europe. The infrastructure implications run deep: scalable yet confined architectures are needed, capable of serving dozens of simultaneous interviews with acceptable latency and demonstrable compliance.

Beyond the hype: what to watch

The funding round tells us investors believe in the marriage of video and AI for hiring. But history shows that real enterprise adoption travels through trust. And trust, in 2025, is built on the guarantee that data never leaves the corporate perimeter. For those evaluating on-premise deployment of similar solutions, the trade-off is not merely economic: it is between the iteration speed of the cloud and the granular control demanded by compliance. AI-RADAR will track these developments, matching startup moves with the actual demands of those who want to bring AI home, not just subscribe to it.