Member of Technical Staff, AI Office

Take AI from the frontier into the field. The AI Office is a small team with a direct mandate from the CEO to make Bittium an AI-native company. We build the AI tools and infrastructure Bittium works with, and work inside our business segments to bring AI into their products. We are hiring two new Members of Technical Staff to join the AI Office and work side by side with our Head of AI. What you build ends up in your colleagues' daily work and in products depended on by defence forces, public authorities and healthcare professionals. You might be the person who - has rewritten their own agent harness more than once. - shipped a full app in a weekend for a team that wasn't yours, and they still use it. - fine-tuned a model and knows whether the eval moving meant anything. - had a new open-weights model running on their own hardware the day it came out, and had an opinion about it by lunch. - was pushed by API pricing, or dreams of personal AI sovereignty, into squeezing a model onto hardware it had no business running on. - has broken a guardrail, and knows why you never rely on just one. - daydreams of dark software factories. What you will do We shape the role around the person, and we make room for exceptional people. If your strength isn't listed below, tell us. - Keep us ahead. You live and breathe the AI takeoff: the models, the hardware, the tools and where the whole industry is heading. Separate what matters from the noise early, and steer where we put our effort. - Go where the work is. Sit with legal, finance, sales, R&D or production, find the work AI should be doing, and build it fast. - Run and grow our AI infrastructure and applications. Inference on our own GPUs, the services on top of it and the end-user applications people work with every day. When a new model is worth having, it is live the same week. - Build the AI our products run on. Build and maintain the agent harnesses, models and safeguards our business segments build products on, and keep them working all the way out to battlefield conditions. - Make models ours. Post-train open-weights models until they beat the general-purpose ones at our work. - Make agents safe to trust. Alignment on the model side, and a hard boundary around it that holds even when the model gets it wrong. What we are looking for above all Taste. You know what good looks like in code, in interfaces and in products, and you don't ship less. You look for the simple solution even when it takes longer to find, because clarity is worth the effort, but you also know when to take a shortcut. An edge. At work or outside it, you have taken something further than most people ever do. We care about that far more than titles, degrees or years of experience. Ownership. We set the direction together, and you own what you build and answer for the results. Lots of freedom, lots of responsibility, so you need to be good at directing your own work. What we ask - You have mastered the craft of working with language models: prompting, context, agent skills, multi-agent orchestration and compiling intent into working software, including the spells that make the difference. - You build real software, fast, with coding agents, and you have opinions about where agentic development is going. One thing is sure: the future is not TUIs on laptops. - A strong full-stack engineer: frontend, APIs, databases, auth and deployment (ours is mostly Go, Python, TypeScript, SolidJS, Svelte and Postgres today, but we don't hire for a stack). You take a tool from an empty repo to production on your own. - Strong software engineering fundamentals: architecture, data modelling, clean interfaces, testing, Linux and networking. When agents write the code, fundamentals are how you judge it. - An eye for opportunity. You have a vision. You spot what should be automated and where a product is hiding, and you find the real problem behind a request, working easily with people who are not engineers. - Fluent English and at least working Finnish. Any of these will make you stand out - Security: agent sandboxing, red or blue team experience. - Evals: designing evals that measure what matters in real work, and knowing when a good number is misleading. - AI products: you have taken an AI-driven product or technology from idea to production, with real users or customers relying on it. - Inference and GPU systems: SGLang or vLLM internals, speculative decoding, disaggregated serving, KV-cache offloading, low-bit quantization, kernels in Triton or CUDA, ROCm. - Infrastructure and trust: confidential computing and attestation, bare metal, AWS or other cloud. Model training: post-training, preference optimization, RL, distillation, datasets, alignment and safety. - AI hardware beyond NVIDIA: AMD, Tenstorrent, Cerebras, Etched, hardwired inference silicon, NPUs and other edge accelerators. - Classic ML and sensing: anomaly detection, RF signal processing, biosignals (ECG, EEG), computer vision, sensor fu