Canopy

Canopy LabsaboutBuilding realtime interactive models for avatarsCanopy Labs is building realtime interactive models for avatars. We think this is one of the most important open problems today. In the future billions of people will interact with these avatars. Here are some of the ways we think society will change:A billion children will be given personalised authentic educational experiencesEvery corporate call will have an avatar, that can add information, contribute superhumanly to meetings naturallyA meaningful fraction of the internet will adopt AI native interfaces for more immersive experiencesmodelsLast year we released Orpheus TTS. Orpheus TTS has been adopted by the largest leading companies in the world, been downloaded over a million times, and forked and fine-tuned thousands of times.Since then we've raised $17M, grown the team, and are building the leading speech and avatar models, which are currently — but not for long — private.jobsGeneral EngineerSan Francisco/LondonData OpsSan Francisco/LondonSalary · $120k+ base with bonusStock · $200k+ in optionsResearchSan Francisco/LondonCreative DirectorLondonSalary · $120k+ with performance based bonusStock · $50–150k in optionsBrand/Media DirectorLondon/SFSalary · $80–120k USDStock · $50–100k in options --- Canopy LabsBack to rolesGeneral EngineerSan Francisco/LondonAbout the roleWe are looking to work with engineers who are interested in building reliable systems for deploying and serving machine learning models in production. A large part of the work involves building infrastructure and services that support ML models in real-world applications.Why you may want to joinWork directly on production systems powering state-of-the-art speech and avatar modelsSmall team where every engineer has outsized impact on the productRapid iteration cycles — your code ships to real users fastDeep exposure to cutting-edge ML research and infrastructure challengesExperienceExperience working with modern ML tooling (e.g. Hugging Face, PyTorch, model serving frameworks)Building production systems around ML models, including data pipelines and inference servicesFamiliarity with infrastructure and DevOps tooling such as Docker, Terraform, and CI/CD systems (e.g. Jenkins or similar)Typical work on the teamDeveloping backend services for model inference (for example using FastAPI or similar frameworks)Working with containerized environments and infrastructure automationBuilding web interfaces or tools (e.g. React-based dashboards or internal tooling) to interact with deployed modelsWhat is the interview process?Step 1: Introduction CallIntroduction call with a team member. The goal of this call is to:Assess your fit from a skill perspectiveAssess your fit from a culture perspectiveGive you a chance to ask questions and find out moreStep 2: Technical InterviewYou will be asked to complete an engineering exercise related to your specialties and skillsets. This could be something like building a data pipeline or optimising a model.Normally it will last about 3 hours:20 minutes of planning10 minutes of discussion2 hours 15 minutes of independent coding and implementation15 minutes of reviewHere is an example coding interview for General Engineer - Data EngineerStep 3: On-siteYou will spend 2–5 days in person with the engineering team. Depending on the situation you would either work through a problem the team has already solved, build something new with the team, or another activity that seems productive with two aims:Helping the engineering team get a sense of working with youHelping you get a sense of working with the engineering team --- Canopy LabsBack to rolesData OpsSan Francisco/LondonSalary · $120k+ base with bonusStock · $200k+ in optionsAbout the roleData is the lifeblood of our models. In this role you will own the end-to-end data pipeline — sourcing, negotiating, and closing large-scale data contracts, then building the operational processes to get that data to the research team quickly and reliably. You will sit at the intersection of the research team, third-party data providers, and customers.Why you may want to joinOwn a critical function — the quality of our data directly determines the quality of our modelsUnusual mix of negotiation, operations, and technical work rarely found in one roleWork with actors, directors, and creative professionals alongside a world-class research teamHigh autonomy to design processes from scratch at a fast-growing companyWhat you will doNegotiate and manage large data licensing and acquisition contractsDesign and run operational workflows that move data from source to training pipelineBuild lightweight internal tooling and automation to scale data operationsCoordinate across research, engineering, and external partners to unblock model developmentRun processes for actors and directors involved in data capture sessionsTypical backgroundsTop STEM programsInvestment banking or private equityManagement consulting (e.g. McKinsey, BCG, Bain)Operations roles at high-growth startupsWhat is the interview process?Step 1: Introduction CallIntroduction call with a team member. The goal of this call is to:Assess your fit from a skill perspectiveAssess your fit from a culture perspectiveGive you a chance to ask questions and find out moreStep 2: Spend Time with the TeamYou will spend time with the team in person, learning about the problems we face across data sourcing, pipeline operations, and partner coordination. You will share your own insights, discuss challenges, and work through real scenarios together.Step 3: On-siteYou will spend 2–5 days in person with the data ops team. Depending on the situation you would either work through an operational challenge the team has already tackled, take on a new data pipeline or vendor coordination problem with the team, or another activity that seems productive with two aims:Helping the data ops team get a sense of working with youHelping you get a sense of working with the data ops team --- Canopy LabsBack to rolesResearchSan Francisco/LondonAbout the roleIn this role you will create new learning techniques and architectures. You will take ideas from new papers and research and investigate promising ones.Why you may want to joinPush the frontier of real-time speech and avatar generationYour research ships — models you build will be used by millions of peopleWork alongside a small, focused team with strong maths and ML backgroundsFreedom to explore novel architectures and training paradigmsHere are some typical profiles of people on the research team, if you feel like you would fit in or have anything additional to contribute we'd love to chat!ML Masters/PhDThe typical profiles of the team currently include:Working at top research labs (i.e. Harvard Robotics Research)Published in top journals (i.e. NeurIPS)So you might fit in well if you find you have similar interestsMaths/PhysicsThe typical profiles of the research team currently are:Scored well on Math/Physics olympiads (for example 20+ BMO2, IPhO)Done well in Maths/Physics degree programs (for example top 10/200)Done impressive research at a good lab perhaps as a MsC or PhDSo you might fit in well if you find you have similar interestsWhat is the interview process?Step 1: Introduction CallIntroduction call with a team member. The goal of this call is to:Assess your fit from a skill perspectiveAssess your fit from a culture perspectiveGive you a chance to ask questions and find out moreStep 2: Technical Interview(s)You will do a Maths and/or ML interview for the research role. You only need to do well enough on one of these interviews.Both interviews are ideally in person.The Maths interview is 90 minutes, with 45 minutes of independent work and 45 minutes of discussing the problems with a team member.Here is an example Maths interviewThe ML interview is 3 hours where you read a paper for 45 minutes, discuss the research with a team member for 15 minutes, then implement the paper in an adjacent task for around 105 minutes, followed by a short 15 minute review at the end.Here is an example ML interviewWe recommend you familiarise yourself with the Hugging Face transformers library and trainer before the ML interview.Step 3: On-siteYou will spend 2–5 days in person with the research team. Depending on the situation you would either work through a problem the team has already solved, work on a new problem with the team, or another activity that seems productive with two aims:Helping the research team get a sense of working with youHelping you get a sense of working with the research team