Research Engineer

We’re building the AI that runs companies.

Sapien is the intelligence layer for enterprise finance. Our agents deeply understand companies across their massive, key financial data sources from Excels to ERPs to CRMs. Why? The world’s largest companies are flying blind. The most crucial financial infrastructure is scarily fragile, and multi-million-dollar decisions are made every day relying on slow and inaccurate analyses.

We’re productizing the cutting edge of research to build systems that go beyond information querying to work alongside finance teams, executing complex workflows and enabling actions that drive growth. Over time, Sapien learns more about how companies operate and builds unique, specific intuitions just as humans do so it can adapt to the most important nuances—becoming what we like to call the "shadow CFO".

We’re solving the hardest technical problems to enable humans to do more of what matters: the high-level strategic decision making, while doing less of what doesn’t: the low-level data gathering and number crunching that dominates human financial work. Sapien is already deployed at manufacturing, software, and services businesses where we’re saving hundreds of hours, understanding models to enable new forms of analysis, and even uncovering multi-million dollar mistakes (the largest so far is $12M+).

Can AI run a company? We’re proving that it can.

To build Sapien, we have to tackle the hardest problems in agent capabilities and verifiability. Your main work will consist of building out our agent backbone and company representations that enable Sapien to reason over complex data, execute verifiable and observable processes, and build deeper intelligence about companies and financials over time. Some key ideas you'd work on include:

• Implementing new agent architectures for observability, HITL, and controlling context

• Employing library learning to reduce LLM dependencies for planning, codegen, and data localization

• Building graphs of data and methods of searching through it efficiently using embeddings and semantic clustering

• Parsing different modalities such as Excel and integrations into a unified schema

• RL and fine-tuning based on user interactions/pairwise data signals

• Benchmarking and evaluation suites to quantify Sapien’s performance

End-to-end thinking with a deeply technical background is essential to succeed in tackling the hard problems we’re working on. Some good signals of this tend to include:

• Strong engineering skills and intuitions—experience solving complex, open-ended problems

• Deep algorithmic thinking, often developed through ML research or competitive programming/math

• Experience building production-grade systems/deployed models at scale

More generally, we’re excited about:

• People with a deep desire to own hard problems across their entire loop, taking challenges all the way from research to product.

• People who love solving new, open-ended problems every single day—adapting fast to new, seemingly impossible settings

• People whose friends would describe them, above all, as curious. People who obsess over the "why" and love pushing their thinking as far as possible.

• People that deeply care about what they’re working on, bringing in new knowledge from papers and engineering blogs they’ve read to engage in thought-provoking discussions

We believe in R&P: research and product. We’re a talent-dense team that aims to excel in all dimensions, pushing bounds of possibility by iterating rapidly. We bring this passion from all previous walks of life, whether it was hacking government facilities, winning top research awards, being top global golfers, blowing up testing liquid-fueled rockets, and much more.

We build everything in-house and are constantly shipping feature-enabling improvements to customers. Our goal is to move fast and iterate relentlessly with customers—gathering data on which features are most useful or accurate, and the way in which users interact with them. Every day, we push both our underlying technical architectures and our product experience to new limits—aiming to always be tech-first, but user-centric.

We’re based in San Francisco, with plans to move to NYC in the coming months to be even closer to customers. We’re a tight-knit, energetic team that love spending long hours nerding out about hard problems, reading papers, or whiteboarding logic puzzles. And we occasionally let loose over poker nights, spikeball matches, sports watch parties, movie nights, hackathons, philosophical debates, among many other events at our office.

Submit your application here and we’ll reach out to chat further if there’s a good fit! Our process consists of a 2-3 chats/technical interviews and a take-home assignment. We then do an in-person work trial to really get to know each other, dig deeper into the technical work we do, and discuss ideas.

Apply to this Job