Software Engineer (Foundational AI)

About The Company The Best of Both Worlds At Maisa, we believe great work happens when people have the freedom to choose the setup that works best for them. Prefer the flexibility of remote work? You’re welcome to work remotely. Most of our roles are currently focused on Spain, although depending on the position and profile, we may be open to hiring from other locations. Looking for some social energy? Our hybrid setup gives you the option to connect and work alongside teammates at our hubs in Valencia, Barcelona, Madrid and Murcia. And as we grow, our spaces will grow with us: when we build a team of more than 10 people in a location, our ambition is to create a hub there too. We also want the time we spend together in person to be worth it. Today, that includes weekly transportation from the city center to our Valencia headquarters and weekly team lunches in Valencia and Barcelona, with more experiences to come as our teams and hubs evolve. The Culture is OURS to Create We’re still shaping how we work together—and that’s part of what makes joining Maisa now specifically exciting. Everyone who joins has the opportunity to influence how we collaborate, connect, and build our culture. Rather than inheriting a finished playbook, you’ll have a voice in creating it. Remote when you want flexibility. Together when you want connection. And a culture you can help shape from day one. About The Role We're looking for a Foundational AI Engineer to join our AI team, working directly with our co-founder and Chief Science Officer. This role is about building our own models. Rather than working on top of existing models, you'll train, fine-tune, post-train, and evaluate the small and large language models that power Maisa Studio — the capability that lets us run the majority of our clients' workloads on our own models and dramatically reduce the cost and dependency of using frontier providers. It's deep, hands-on model work in Python and PyTorch, with a strong emphasis on evaluation and on getting models into reliable production use inside highly regulated environments — banks, insurers, and telecoms where getting it wrong isn't an option. This is an early-stage company, and the AI team is growing quickly. You'll have real ownership over what we build and how we build it. What You'll Do - Train, fine-tune, and post-train Maisa's own small and large language models — from data preparation through to evaluation and release - Design and run rigorous model evaluation: benchmarks, regression sets, and quality criteria that reflect real client use cases rather than leaderboard scores - Build and improve training pipelines and infrastructure, including data curation, experiment tracking, and reproducibility - Apply and adapt modern training techniques — parameter-efficient fine-tuning, post-training and alignment methods — and evaluate their real impact - Work with our AI infrastructure function on model serving and inference optimisation (vLLM, SGLang), including the constraints of deploying inside client environments - Monitor deployed models for quality degradation and drift, and drive the retraining or corrective work required - Help shape our model strategy — where custom models genuinely add value versus using frontier models or deterministic code - Collaborate closely with our Applied AI, product, and engineering teams to make sure models meet real product and client needs What We're Looking For - Genuine, current, hands-on experience training and fine-tuning models — this is a builder's role, not a supervisory one - Strong PyTorch depth: comfortable building and modifying model architectures directly, not only running existing pipelines - Rigorous approach to evaluation — you can explain how you know a model is actually good, and how you'd detect regressions and degradation - Solid software engineering fundamentals in Python - Sound judgment on when custom models are worth the investment, weighed against accuracy, cost, and latency - Comfortable working in a fast-moving, ambiguous environment with a high degree of ownership - Strong communication skills — you'll work closely with engineers, product, and leadership Nice to have - Experience with model serving and inference optimisation at scale (vLLM, SGLang, quantisation) - Experience deploying models in regulated, on-premise, or air-gapped environments - Familiarity with the Hugging Face ecosystem (transformers, datasets, PEFT, TRL) - Experience with distributed or multi-GPU training - Open-source contributions or published work in the ML community - Background in an early-stage or scaling startup Location Fully remote. We have offices in Madrid, Barcelona, and Valencia — visiting is entirely optional. Why join Maisa! You'll be joining one of Europe's most exciting early-stage AI start-ups, working on genuinely differentiated technology: building and training our own models rather than wrapping someone else's, in a domain where reliability and auditability actually matter. You'll work directly alongside our co-founder and Chief Science Officer — a prolific and well-known contributor in the open-source ML community — and have real influence over the direction of our model strategy. Following our recent $25m Seed Round, backed by leading Venture Capital firms including Creandum, Forgepoint, NFX, and Village Global, we're scaling quickly and realising significant enterprise traction — working with clients including Santander. This is your opportunity to solve real enterprise AI challenges alongside deep technical and industry experts, where you'll be challenged daily and accelerate your learning and development. By submitting this application, I agree that my personal data will be collected, processed, and retained by the company solely for the purposes of managing and assessing my candidacy.