Senior Software Engineer, Tavus

• You're proactive. When you see an issue, you fix it. Nobody has to ask
• You're autonomous. Given a loosely defined project, you figure out what "done" means, make the judgment calls, and deliver
• You're energized by unfamiliar problems. A lot of the work here has no playbook, and you'll take on projects outside your direct experience
• You adapt as priorities evolve. This space moves fast, and the most important thing to build can change as we learn. When it does, you adjust and keep moving
• You hold a high bar. What matters to you is how the product feels to the person on the other end of the conversation. Clean code and clever architecture are means to that end
• You're passionate about the future of interfaces. How people and computers interact is being reinvented right now, and you want a hand in shaping it
• Fluent in Python, with real depth in concurrent programs. Low-level concepts (asyncio, multiprocessing, event loops, the GIL, shared memory) are dear friends, and you design with them in mind
• You've built real-time systems, whether video, audio, or model inference, and you know what a millisecond latency budget feels like
• You bring a senior track record: you set direction, own decisions, and get things over the finish line
• You explain complex things clearly, and you use that to teach the people around you
• If you don't check every box but this sounds like the work you want to be doing, apply anyway

Desirable

• We're hiring a Senior Software Engineer to be a technical driver on Tavus's Conversational Video Interface, our multimodal real-time conversational product. It's a hands-on role. Your code runs in the middle of live conversations between people and PALs
• Ship features end-to-end. You'll take product priorities that start out loosely defined, like voice localization for accent support or multilingual sentence endpointing, and turn them into a spec, an architecture, and shipped features
• Partner with Research to bring state-of-the-art real-time models into CVI. Sometimes that means dipping into research territory yourself
• Fix what you find. If you spot a flaky process or a slow path, you don't need a ticket or anyone's permission to go after it
• What this role has shipped:
• Guided Browser Control: a framework that lets PALs drive a real web browser mid-conversation, narrating each click and scroll out loud
• Magic Canvas: interactive cards rendered right on the video stream. The PAL puts the right card up at the right moment: a question, a text box, a calendar, a chart, an alert
• Memories and Knowledge Base: the context layer that lets PALs recall past conversations and pull from document stores in real time, without breaking stride
• Utterance-to-utterance optimization: shaving hundreds of milliseconds off conversational turns