Our investment in Parsed

We were initially introduced to this incredible team by Tom Kelly, CEO of one of our fastest-growing healthcare portfolio companies, Heidi, and we jumped at the opportunity to get to know them. Tom was raving about Parsed’s ability to help Heidi train auditable models on its proprietary data, directly addressing a critical challenge for Heidi.

We’re proud to lead Parsed’s $3.7m first round of funding, alongside leading angels including Thomas Wolf, co-founder of HuggingFace, Mehdi Ghissassi, former director and head of product at Google DeepMind, and Lord David Prior, ex-chair of the NHS. The round will accelerate Parsed’s mission to build the AI post-training platform for customers to train custom frontier models that outperform big-lab generalists — with minimal engineering lift. Their position piece is here.

Parsed is building AI differently. Most companies today rent generalist models that don’t learn. Parsed enables companies to own specialist models that improve continuously and can be owned by customers. They help customers build expert evaluations that mirror subject-matter experts and apply reinforcement learning to reach state-of-the-art performance. The results are models that outperform the largest models from OpenAI, Anthropic, and Google at a fraction of the cost with almost no engineering overhead.

Crucially, Parsed integrates mechanistic interpretability tooling, making models explainable and auditable. In industries like healthcare and law, attribution and transparency are essential for deployments at scale. By embedding reliability and accountability into AI, Parsed makes deployment in high-stakes workflows not just possible, but safe.

As always, it’s the Founders we’re proud to partner with.

• Mudith Jayasekara: Rhodes Scholar, Oxford PhD candidate in engineering, medical doctor, bridging clinical insight with technical depth (and an ex-elite pole vaulter for Australia!)
• Charles O’Neill: Researcher in mechanistic interpretability and reinforcement learning, with experience across Stanford, Johns Hopkins, MATS (ML Alignment and Theory Scholarship), and NASA.
• Max Kirkby: Rhodes Scholar, Oxford PhD candidate in computational neuroscience.

Together, they combine scientific rigour, technical expertise, and relentless execution.

Parsed is tackling a $6B+ and rapidly growing market defined by enterprise API spend on large language models. Their first wedge, healthcare scribes, a high-volume, high-stakes use case where owning and optimising models creates massive value. From there, Parsed is positioned to become the infrastructure layer for domain-specific AI, powering industries where accuracy, efficiency, and explainability are non-negotiable.

We couldn’t be more excited to partner with Mudith, Charles, Max, and the team as they build the next generation of interpretable, domain-specific AI infrastructure. The need is urgent, the market is vast, and this team is uniquely equipped to lead it.