Sr. Staff Engineer (Conversational/Voice AI)

Sr. Staff Engineer (Conversational/Voice AI) Department: Engineering Team: Machine Learning Location: San Francisco, United States Type: Full-Time **About the Role** Uber’s Customer Obsession team builds the platform and AI that powers world‑class support across **mobile, web, and voice** at global scale. We are now hiring a Senior Staff Engineer to **architect, productionize, and scale an autonomous support agent** that resolves customer issues end‑to‑end. Experience with **voice agents** and **agentic architectures** is a major plus. You’ll push the state of the art in GenAI for customer service—LLM orchestration, evaluation, safety guardrails, multilingual support, and real‑time voice—while holding a very high bar for reliability and cost efficiency. We are still at an early stage and value candidates with bias for action who get creative with GenAI tools to accelerate execution and experimentation. **What the Candidate Will Need / Bonus Points** \-\-\-\- What the Candidate Will Do ---- 1. **Own the end‑to‑end agent architecture**: agentic planning and execution loops, long-term memory, persona/voice, knowledge routing, and policy enforcement for compliant, on‑brand conversations. 2. **Ship production systems** that handle millions of conversations with rigorous **SLOs, fallbacks, and canaries**; design graceful degradation (e.g., human handoff) and safety guardrails (prompt‑injection, jailbreak, PII redaction). 3. **Lead voice agent initiatives**: Drive the development of Uber’s voice support agent—covering real-time speech recognition (ASR), text-to-speech, natural turn-taking (barge-in and endpointing), and reliable telephony/WebRTC integration. Ensure low-latency, high-quality interactions that remain robust even in noisy environments. 4. **Advance retrieval & reasoning**: Build next-generation retrieval and reasoning pipelines, where the agent can search across different knowledge sources, apply policy-driven tools, and call structured workflows and ensure that responses are consistently grounded. 5. **Establish evals that matter**: offline rubrics, simulated scenarios, safety tests, cost/latency t