Skip to content

Research

Research That Ends Up
in Production

A home for people who want to understand how AI systems behave, not just build with them. Two programs, one shared habit: test the claim before anyone relies on it.

Open to US-based candidates only

Why We Do Research

The Hard Part Is Reliability

We build AI software, media and investing systems for clients who need them to work on Monday morning. Demos are easy. What is hard is an agent that behaves the same way the hundredth time, a model that stays private, and a result that holds up outside the benchmark it was tuned on.

Our research exists to answer those questions before a client has to. Findings go straight into what we build and recommend, and the best of them are published on our Papers and Blog.

Programs

Two Ways to Join

Both programs are open only to candidates who live in the United States and are authorized to work in the US.

Internship 3 to 6 months Austin, TX / Remote (USA) US candidates only

Research Internship

For undergraduate and graduate students, and early-career researchers, who are serious about AI research. You work on a project matched to your interests and our roadmap, contribute to work already in flight, and leave with a result you can put your name to.

What you'll do

  • ✦ Own a research project aligned with the lab roadmap
  • ✦ Contribute to ongoing projects, not side quests
  • ✦ Produce publication-track results or research artifacts
  • ✦ Present your findings to the lab at the end

What you get

  • ✦ Weekly 1:1 mentorship from a senior researcher
  • ✦ Full access to lab compute and infrastructure
  • ✦ A stipend scaled to duration and contribution
  • ✦ Strong consideration for full-time roles afterwards

Eligibility: you must live in the United States and be authorized to work in the US. Undergraduate, graduate or early-career; Python and a habit of reading papers expected.

Timing: no fixed cohorts. We hire on a rolling basis, and each internship runs 3 to 6 months from the start date we agree.

Apply Now
Residency 6 weeks Remote, in person (Austin, TX) or hybrid US candidates only

Research Residency

For curious minds with a specific question about AI, whatever their background. No formal research credentials required: we care about intellectual seriousness, and some of the best questions come from outside computer science. Kept deliberately small, two to four residents per cohort.

What you'll do

  • ✦ Explore a research question you define, in protected time
  • ✦ Join daily seminars, working groups and paper readings
  • ✦ Work alongside our researchers and engineers
  • ✦ Close with a showcase of what you found

What you get

  • ✦ Six weeks of protected exploration time
  • ✦ A personal research mentor
  • ✦ Remote, in person in Austin or hybrid: your choice
  • ✦ An optional path into the Research Internship

Eligibility: you must live in the United States and be authorized to work in the US. Any background; bring a question, not just an interest.

Timing: rolling, with flexible start dates for the right resident.

Apply Now

At a Glance

Research Internship Research Residency
Length3 to 6 months6 weeks
WhereAustin, TX or remote within the USRemote, in person in Austin, TX, or hybrid
WhoStudents and early-career researchersAnyone with a sharp question
SupportStipend, mentorship, computePersonal research mentor
OutcomePublication-track resultShowcase to the lab
EligibilityUS-based candidates only, authorized to work in the US

Research Directions

Three Questions We Keep Asking

Bring your own question inside one of these, or one we have not thought of yet.

Reliable Agents & Evaluation

How do you know an agent is better, not just different? Benchmarks reward the test; clients need the hundredth run to match the first.

  • ✦ Evaluations that predict production behaviour
  • ✦ Failure modes of tool use and long tasks
  • ✦ Guardrails that live in code, not in prompts

Private & Efficient AI

What can run on hardware you own, with data that never leaves it? The question behind Sepia, and behind every client who cannot send data to a third party.

  • ✦ Small models that hold up on real tasks
  • ✦ Cost, latency and quality trade-offs in serving
  • ✦ Privacy guarantees you can actually check

AI for Markets & Media

Where do models add real signal, and where do they only sound confident? From alpha without prediction to generative media that changes what a production costs.

  • ✦ Agentic decision systems with honest evaluation
  • ✦ Backtests that survive contact with the market
  • ✦ Generative media quality, control and cost

Life in the Lab

How the Work Happens

Mentorship

A named mentor and a weekly 1:1. You are never left guessing whether you are on track.

Reading Group

Papers read closely and argued about, often the ones that end up summarized on our Papers page.

Compute

The same infrastructure our engineers use. Run the experiment rather than argue about it.

Artifacts Over Slides

Code, writing, experiments and models. Something that runs is worth more than a deck about it.

Close to Production

Findings reach engineering in days, not quarters. You will see where your work lands.

Publish

Good results get written up and published with your name on them.

Who We Look For

Trajectory Over Credentials

Computer science, mathematics, cognitive science, economics, linguistics, philosophy: the backgrounds vary. The patterns do not.

  1. 01

    You arrive with a question, a specific one, not just a general interest in AI.

  2. 02

    You read across fields and connect ideas that do not usually meet.

  3. 03

    You are comfortable not knowing for a while; foundational work is slow before it is fast.

  4. 04

    You build things: code, essays, experiments, models. Show us one.

  5. 05

    You are getting better quickly, and can show us how.

How to Apply

Send us a short note through the contact form and choose Research programs. Tell us which program, the question you want to explore, why now, when you could start, and link to anything you have built or written. No CV template required.

We review on a rolling basis. Both programs are open only to candidates who live in the United States and are authorized to work in the US.

Apply to a Research Program

Looking for a job rather than a program? See open roles and how we interview.