Careers at Keen Research

We're building the future of voice technology: faster, leaner, and completely offline.

Big Tech companies made voice assistants creepy and bloated. We're taking voice back to where it belongs: fast, private, on-device, and embedded directly into the next generation of intelligent apps and hardware.

Our SDK helps developers create speech-enabled products without the cloud. That means no latency, no privacy compromises, and no internet required. This isn't a toy; it's powering tools that help kids learn how to read, disabled people control devices with their voice, and enterprises build specialized voice interfaces that Big Tech companies would never touch.

We're profitable, battle-tested, and bootstrapped by design. Our customers include startups, public companies, and global brands who are already building with our tools.

Now, we're ready to scale. We're looking for people who move fast, think clearly, and want to own part of something real.

We don't do politics. We ship fast, learn faster, and obsess over quality.

If you're hungry to grow, not afraid to build what's never been built, and want to work with a tight team that gives a damn, then this is your moment.

Join us. Be early. Build the future.

Open roles:

Principal ASR Engineer

The role is based in Sausalito, California; we will consider remote candidates.

We're looking for a Principal ASR Engineer to own and ship core parts of our offline voice platform.

This is a hands-on, high-ownership role. You'll work across the system, from models to data to APIs to deployment, and directly shape how we build and evolve our speech platform.

This is a build-and-ship role with real impact. Most successful candidates will have several years of experience building and shipping ASR systems, but we also care deeply about how you work. If you're early in your career (for example, just out of a strong PhD program) but have high agency, move fast, and take ownership, we want to hear from you.

What You'll Do:

  • Own the modeling pipeline end-to-end: training, evaluation, and deployment across devices
  • Build ASR systems for real-world use cases such as children's speech, domain-specific vocabularies, literacy instruction and assessment
  • Design and improve data pipelines: curation, labeling strategies, and evaluation frameworks
  • Use synthetic data and augmentation where it makes sense
  • Work under real on-device constraints: latency, memory, and accuracy tradeoffs
  • Own parts of the SDK end-to-end: from API design to model deployment across platforms
  • Ship continuously, from small fixes to larger features, without unnecessary complexity
  • Debug and improve systems in production

Who You Are:

  • You have strong ASR fundamentals and experience with modern ASR approaches such as CTC, RNN-T, Whisper-style, or similar systems
  • You understand the full lifecycle: data → training → evaluation → deployment
  • You're comfortable working with messy, real-world datasets and treat data as a first-class part of the system
  • You understand when and how to use real vs synthetic data effectively
  • You implement well-scoped tasks quickly and correctly, and default to simple, working solutions over complex ones
  • You can scope and deliver larger features when needed without overengineering
  • You know when to move forward and when to align, and you don't spend hours going down the wrong path without resetting
  • You can quickly understand unfamiliar codebases and make minimal, precise changes instead of broad rewrites
  • You are comfortable using AI/LLM tools to accelerate development, but you critically evaluate, adapt, and simplify generated solutions rather than following them blindly
  • You want to own what you build and are excited by a small team with high leverage and high responsibility

Your Toolkit Likely Includes:

  • C++, Python
  • PyTorch, ESPnet, Kaldi, OpenFst, or similar tools and frameworks
  • Experience with on-device or performance-constrained systems
  • Nice to have: build systems, Android, iOS, ObjC, Java, JavaScript, low-level optimization

Our Technical Approach:

Our current core includes Kaldi/OpenFst because it still performs extremely well for specific real-world use cases where control, determinism, and footprint matter. At the same time, we're actively evolving toward more end-to-end solutions and more capable model-driven systems. We're looking for someone pragmatic enough to work effectively with the current system, while helping us push toward a state-of-the-art on-device voice solution. Our roadmap includes small language models (SLMs), dialog systems, and more capable on-device language understanding.

Why Join:

  • Founding-level ownership of product-critical systems
  • Direct influence over model design, data strategy, and system architecture
  • Work on problems with real human impact: children's literacy, accessibility, pronunciation, and on-device constraints
  • Build a fully offline voice platform, a fundamentally different challenge from cloud AI
  • Profitable, bootstrapped company with real customers
  • Zero bureaucracy, zero BS, just smart people solving real problems

Don't tick every box? If you care deeply about building real speech systems, move fast, and take ownership, let's talk.

Apply now