Operators who became investors.
We built GPU scheduling layers, model serving platforms, and low-latency execution systems before Firntal. The thesis came from that work — not from watching the category from the outside.
PhD in distributed systems from ETH Zurich. Spent six years as a principal engineer at a hyperscale cloud provider in Zurich building the GPU workload scheduling layer — the infrastructure that sits beneath every training run and inference request. The work meant understanding, at a kernel level, why GPU utilization drops, where batching breaks down under load, and what happens to job latency when the scheduler makes the wrong queuing decision.
Co-founded an ML inference startup in 2018 that was acqui-hired in 2020. Left with a clear conviction: inference cost and p99 latency would determine which AI applications survive the transition from demo to production. Founded Firntal in 2021 — before inference infrastructure was the obvious thesis — and has led every investment decision since. Lukas holds board seats or observer roles across the portfolio.
Eight years in infrastructure engineering — built the model serving platform at a European unicorn AI company, then led platform engineering at a Zurich-based MLOps startup through a growth stage that required rearchitecting the entire serving stack. The work covered KV cache management, tensor parallelism across multi-GPU nodes, autoscaling for latency-sensitive inference endpoints, and the operational complexity that accumulates when you try to serve 40+ models from one platform.
Joined Firntal in 2022. Sarah leads technical diligence on every new investment — reviewing system architecture, serving design, hardware assumptions, and scaling models with founding teams directly. She also runs the bi-weekly Architecture Office Hours program: working sessions with portfolio CTOs where real system design decisions get worked through, not summarized for a slide deck.
Former quantitative researcher at a Swiss systematic trading firm, where he spent four years optimizing low-latency order execution systems — co-location, order book modeling, microsecond-level latency reduction. The discipline maps directly to inference performance engineering: both are problems of minimizing tail latency under throughput pressure, with hardware constraints you cannot fully control.
Transitioned to venture through an ETH-affiliated accelerator program in 2023. Leads financial modeling, fund operations, and LP relations at Firntal. Runs the portfolio fundraising preparation program — financial model construction, investor targeting, and narrative positioning — starting 9 months before portfolio companies need capital, not in the final weeks of a runway crunch.
What the term sheet is actually a bet on.
Sarah holds bi-weekly working sessions with portfolio CTOs — not advisory calls. Founders bring real architectural decisions: batching strategy under bursty load, KV cache sizing for long-context workloads, hardware selection for latency-critical serving. Sarah brings the context of having built and maintained production serving platforms at scale. The sessions produce actual decisions, not slide commentary.
The ETH Zurich ML engineering community runs through Firntal. We connect portfolio companies with senior ML engineers, infrastructure leads, and potential technical co-founders from the networks we built as practitioners.
Niklas maps infrastructure buyers across financial services, healthcare, and manufacturing — sectors where AI inference is moving from pilot to procurement. We make introductions to engineering directors and CTO-level buyers who are actively evaluating inference infrastructure vendors, not generic LinkedIn connections.
Niklas starts fundraising preparation 9 months before portfolio companies need capital — financial model construction, investor mapping, and narrative positioning. Built with time to iterate properly. Not assembled in the two weeks before the runway conversation becomes urgent.
Our LP base includes corporate strategics with active AI procurement programs. When portfolio companies reach distribution-readiness, we facilitate structured introductions — not just warm emails into procurement black holes.