AI Engineer, Voice and Realtime

Obble; United States; Remote; USD 120,000 - 180,000; Full time

Job summary

Obble is seeking a software engineer to own the realtime AI interview pipeline behind live candidate conversations. You will handle audio ingestion, speech recognition, language-model decisions, speech output, and vendor integrations with careful attention to latency and user experience.

You will ship production systems, design robust prompts, and ensure the system remains fast, reliable, and fair for interviewees across scenarios.

Qualifications

• Strong Python skills with production experience beyond notebooks.
• Experience shipping a production system that calls a language model.
• Practical experience with asynchronous programming and streaming data.
• Measurement-oriented approach to AI quality and using evidence for changes.
• Ability to work across prompts, application code, vendor integrations, and evaluation tooling.

Responsibilities

• Build and operate the realtime interview worker within a latency budget.
• Create prompts as versioned engineering artifacts and evaluate behavior.
• Design scenarios, inspect transcripts, and quantify failures before releases.
• Maintain abstraction over speech and model providers to allow vendor changes.
• Debug asynchronous and streaming systems and improve reliability and observability.
• Focus on candidate experience: clarity, timing, consistency, and fairness.

Skills

Python

Asynchronous programming

Streaming data

Debugging large systems

Tools

WebRTC

LiveKit

Job description

Role overview

Own the realtime AI interview pipeline behind live candidate conversations. The work covers audio ingestion, speech recognition, language-model decisions, speech output, turn-taking, vendor integrations, prompt design, and evaluation—while keeping responses fast enough to feel natural and respectful to the person being interviewed.

Responsibilities

• Build and operate the realtime interview worker, coordinating audio handling, speech-to-text, model calls, and conversational turn-taking within a defined latency budget.
• Create prompts as versioned engineering artifacts and measure their behavior against a stable evaluation corpus.
• Design scenarios, inspect transcripts, analyze failures, and establish quantitative evidence before releasing changes to the interviewer.
• Maintain an abstraction layer around speech and model providers so vendors can change without forcing product-level rewrites.
• Debug asynchronous and streaming systems in production and improve their reliability and observability.
• Consider the candidate experience throughout the system, including clarity, timing, consistency, and fairness of the interaction.

Requirements

• Strong Python skills and experience shipping a production system that calls a language model, beyond notebook experimentation.
• Practical experience with asynchronous programming, streaming data, and the debugging challenges those systems create.
• A measurement-oriented approach to AI quality: you investigate failures systematically and use evidence to support changes.
• Ability to work across prompts, application code, vendor integrations, and evaluation tooling.
• Careful judgment about the people affected by automated interviews and the quality of their experience.

Nice to have

• Experience with WebRTC, LiveKit, or another realtime audio platform.
• Previous work running model evaluations, A/B comparisons, transcript analysis, or other structured experiments on AI output.