AI Engineer - Conversational AI

You’ll own a conversational AI voice assistant that’s actually talking to real customers - answering live inbound calls for disability and absence management.

Callers speak naturally. The assistant verifies who they are. Then it pulls from a knowledge base and gives them answers on the spot.

This isn’t R&D. This is build, ship, and keep it rock-solid in production.

You’ll support the live pilot, drive the shift to a fully agentic design with tighter latency, and harden it for scale and new features through a real engineering and release process.

If you’ve shipped LLM-powered apps into production and kept them running, we want to talk to you.

What You’ll Work On

You’ll be the hands-on engineer owning the full stack of this AI voice experience:

• Conversational AI for inbound customer calls
• Identity verification and secure Q&A flows
• Hybrid scripted + AI flows evolving to fully agentic conversations
• Latency optimization for real-time voice interactions
• Automated evaluation of LLM quality and safety
• Production monitoring, incident response, and cost control
• Regulated data handling with auditability and guardrails
• Multi-environment CI/CD and change management

Under the hood you’ll work with Amazon Connect, Amazon Lex V2, Amazon Bedrock, ElevenLabs, Python, and core AWS services.

Your mission: keep the voice assistant fast, accurate, and safe while you evolve it into a truly agentic system., * Supporting the live pilot by triaging and resolving production incidents and support tickets

• Reviewing call transcripts and tuning prompts, intents, and fallback handling based on real conversations
• Leading the redesign from a hybrid scripted/AI flow to a fully agentic conversational experience
• Driving measurable latency reductions across the voice interaction
• Building test sets and automated evaluation suites to measure answer accuracy, hallucination, and response time
• Running evaluation gates before every release to protect quality
• Implementing safety and compliance guardrails, including PII/PHI protection, redaction, and prompt-injection defenses
• Ensuring robust audit logging and alignment with formal change-control processes
• Delivering through Azure DevOps with stories, acceptance criteria, pull requests, and code reviews
• Managing multi-stage pipelines that promote bots and contact flows from Dev to QA to UAT to Prod
• Creating monitoring dashboards, alerts, and incident playbooks for the AI assistant
• Tracking and optimizing AI token usage and Amazon Connect consumption costs

• Hands-on production AI experience delivering at least one LLM-powered application to real users
• Comfort with prompt engineering, tool/function calling, structured output, and RAG
• Proven work with Amazon Connect and Amazon Lex V2, including contact flows, Lambda integrations, intents, slots, and fallbacks
• Experience with Amazon Bedrock or a comparable managed LLM service (knowledge bases, agents, guardrails)
• Background building evaluation suites for LLM output and tracking accuracy and latency over time
• Strong Python skills plus working knowledge of AWS Lambda, IAM, DynamoDB, S3, CloudWatch, and KMS
• Experience handling sensitive personal data in a regulated environment (insurance, healthcare, financial services, or similar)
• Comfort operating under formal change control, audit trails, and separation of duties

The Experience That Will Really Get Our Attention

You’ve been the go-to engineer for a production conversational system, not just a hackathon bot. You understand where scripted flows end and agentic architectures shine, and you’ve cut real latency out of voice or chat experiences.

Ideally you’ve mixed AI, telephony, and DevOps: building with Connect and Lex, wiring Bedrock or similar LLM services, and shipping through pipelines with hard quality gates in a regulated setting.

High-signal experience: ElevenLabs (TTS, Agents, APIs) * real-time voice latency tuning * Bedrock Agents or Flows * Step Functions * EventBridge * SQS/SNS * Terraform/CDK/CloudFormation for Connect/Lex promotion * Azure DevOps (Repos, YAML Pipelines, Boards) * Java/Spring Boot * TypeScript/Node * Amazon Contact Lens * Amazon Q in Connect * CTR streaming * Kinesis * Lex generative features * multi-language bots * OpenSearch Serverless * Aurora pgvector * Datadog * SonarCloud * unit testing with mocked LLM calls * secrets management * insurance/disability/absence domains, Work hours must align with US Eastern business hours, or have significant overlap with US Eastern mornings through 12 PM ET.