Staff AI Engineer, Conversational Agents

AssistCall, a product of Starling Global, builds AI receptionist and business calling software. The product connects live conversations with business information, appointments and human handoffs. In this role, you will improve the reliability, quality and operation of the systems behind those conversations.

THE OPPORTUNITY

You will lead the engineering of agent behavior across real-time conversations, tool execution, retrieval and human handoff. This is a hands-on staff role for someone who can turn an ambiguous product problem into a measured system, explain its failure modes, and carry it through production. The work includes distinguishing dialogue failures from telephony failures, controlling side effects, preserving tenant boundaries and measuring task completion beyond a convincing demo.

INITIAL PRIORITIES

• Establish a baseline for answer latency, interruption handling, task completion, incorrect tool execution and cost per successful call.

• Ship a bounded workflow with explicit action permissions, confirmation rules, idempotent execution and a recoverable human handoff.

• Build replayable evaluations and release criteria, then document a production incident or failed experiment with the evidence behind the fix.

TECHNICAL CONTEXT

Relevant experience includes real-time model APIs, speech pipelines, retrieval systems, typed tool execution, event-driven services and observability. Experience with WebRTC, SIP or telephony is useful. We evaluate architecture choices through measurements rather than assuming one vendor or orchestration framework is always appropriate.

COMPENSATION

US$220,000–285,000 per year (base salary). This is the expected base salary range for this role. An offer within the range is based on relevant skills, experience, the scope of the role and internal equity. Any separately offered bonus, equity or benefits are not included in this base salary range.

WORK ARRANGEMENT

Remote within the United States. Full-time employee role. Working hours and any role-specific operational coverage will be discussed during hiring.

HOW WE HIRE

Apply to this role through Indeed. Our hiring team can invite you into the AssistCall application process after reviewing your application. Share a résumé and a concise account of relevant work; public project, publication and portfolio links are optional. You can choose a recorded 15-minute technical conversation with Ava, our AI interviewer, or request a human interview. Ava collects and transcribes answers for people to review; these technical roles do not use automated hiring scores, rankings or recommendations. Human reviewers determine next steps, which may include a project discussion, system-design conversation or bounded role-relevant exercise. Do not disclose confidential employer material or customer data.

EQUAL OPPORTUNITY AND ACCESSIBILITY

AssistCall considers applicants based on job-related qualifications without unlawful discrimination based on race, color, religion, sex, pregnancy, sexual orientation, gender identity, national origin, age, disability, veteran status or other status protected by applicable law. Reasonable accommodations are available in the application process as required by law. Contact [email protected] for hiring support, an alternative interview format or accommodation; no diagnosis is required in the initial request. We do not request salary history. See https://career.assistcall.ai/hiring-process and https://assistcall.ai/privacy for information about application processing.

RESPONSIBILITIES

• Design dialogue state, structured tool contracts, retrieval boundaries and durable workflow state for live voice agents.

• Implement cancellation, retries, timeouts and idempotency across model requests, telephony events and external business APIs.

• Build evaluation sets from consented, appropriately handled examples; prevent train/test leakage and track scenario coverage.

• Separate model reasoning from execution authority, including confirmation before payments, account changes and other consequential actions.

• Diagnose latency across speech detection, model inference, tool calls and audio playback using traces and reproducible experiments.

• Lead design reviews and mentor engineers while continuing to write and operate production code.

QUALIFICATIONS

• Deep production experience building LLM applications or conversational agents with measurable reliability requirements.

• Strong Python or TypeScript engineeri…