Research and improve real-time conversational AI for voice drive-thru ordering. Responsibilities include advancing speech, intent, turn-taking, voice activity detection, and order-accuracy models; curating and generating data for rare scenarios; building statistically rigorous benchmarks; evaluating audio and conversational quality under real-time constraints; and translating research into production improvements with applied AI and backend teams.
About the Role
At Arc, we are on a mission to amplify employees at local businesses using AI. With years of experience building and scaling commerce tools for businesses—from farmers markets and neighborhood coffee shops to global retailers, restaurant chains, and major sports venues—we know how to deliver mission-critical technology that integrates easily, delights customers, and works every time.
Voice drive-thru is our first product on that journey: an AI voice agent that takes real orders at the drive-thru speaker, in real time, integrating directly with a business's existing POS and hardware. The hard part isn't just "can an LLM take an order" — it's making the whole conversational loop (does the customer sound done talking, was the transcript actually right, did the model behave consistently across a thousand near-identical calls) hold up against real noise, real accents, real mumbled orders, and a live customer who won't repeat themselves.
As an AI Researcher, you'll push on the modeling and methodology underneath that loop: turn taking, voice activity detection, speech accuracy, order accuracy, detecting side conversations, and the rigorous experimentation needed to know — with statistical confidence, not vibes — whether a change actually made things better. This is a research role in service of a product that's live in real drive-thru lanes today, not a lab role disconnected from production.
What You'll Do
- Improve the models that let the agent understand natural, real-world speech, understand their intent and converse with them naturally while ringing their order in realtime
- Own speech accuracy as an ongoing research problem
- Data modeling to test and train the system against situations that are rare or hard to collect from live calls
- Build benchmarking practices that give the team confidence a change actually helped, rather than just seeming better
- Evaluate models and approaches for both conversation quality and the underlying audio pipeline, balancing accuracy against real-time constraints
- Work with the applied AI and backend teams to turn research findings into real product improvements
What We're Looking For
- Strong applied ML/research background (industry or research lab), with hands-on experience taking models from experiment to production, not just publication
- Experience with speech/audio ML
- Solid Python skills and comfort working with training ML models from scratch
- Experience designing rigorous evaluation methodology — statistically sound benchmarking, synthetic data generation, or dataset curation — for non-deterministic systems.
- Comfort operating with ambiguity — you can take a fuzzy quality problem ("the agent barges in too much") and turn it into a measurable research question
- Strong execution from research idea to real production deployment
Why Join Us
You'll have significant influence over both the product and the research culture at an early-stage company solving a genuinely hard, high-stakes problem: real-time AI that talks to real customers, in service of a mission to help local businesses run better. Voice drive-thru is just the beginning — your decisions will help shape what comes next.
Similar Jobs
Agency • Artificial Intelligence • Blockchain • Web3
Research, design, and train foundation models for robotics; align text-based reasoning with low-level control policies; evaluate zero-shot generalization; run experiments; integrate models into robotic systems and present findings externally.
Top Skills:
PythonPyTorchRt-XTensorFlow
Security • Software
Research AI threats, analyze threat actor behavior, produce actionable intelligence, collaborate with teams, and publish findings in the cybersecurity domain.
Top Skills:
AILlm ApisPythonSecurity Data LakesSIEMSoar
Artificial Intelligence • Information Technology • Software
The role involves applying diffusion techniques and developing AI models to enhance 3D reconstruction pipelines for various industries.
Top Skills:
Computer VisionDeep Learning FrameworksDiffusion ModelsFlow-Based ModelsGanGenerative AiPyTorchTensorFlowVae
What you need to know about the Seattle Tech Scene
Home to tech titans like Microsoft and Amazon, Seattle punches far above its weight in innovation. But its surrounding mountains, sprinkled with world-famous hiking trails and climbing routes, make the city a destination for outdoorsy types as well. Established as a logging town before shifting to shipbuilding and logistics, the Emerald City is now known for its contributions to aerospace, software, biotech and cloud computing. And its status as a thriving tech ecosystem is attracting out-of-town companies looking to establish new tech and engineering hubs.
Key Facts About Seattle Tech
- Number of Tech Workers: 287,000; 13% of overall workforce (2024 CompTIA survey)
- Major Tech Employers: Amazon, Microsoft, Meta, Google
- Key Industries: Artificial intelligence, cloud computing, software, biotechnology, game development
- Funding Landscape: $3.1 billion in venture capital funding in 2024 (Pitchbook)
- Notable Investors: Madrona, Fuse, Tola, Maveron
- Research Centers and Universities: University of Washington, Seattle University, Seattle Pacific University, Allen Institute for Brain Science, Bill & Melinda Gates Foundation, Seattle Children’s Research Institute
