Forward Deployed AI Engineer

About us
Next Dialog automates business communication with AI. Our omnichannel platform handles customer conversations across phone, SMS and messaging: answering questions, booking appointments, qualifying leads and syncing everything to the CRM, 24/7. Our clients range from HVAC installers and real estate brokerages to dental clinics, logistics and energy companies across Poland, Germany, the rest of Europe and the US. We're a small, product-focused team working at the intersection of AI and communication automation, and we're hiring a Forward Deployed AI Engineer to own our AI systems from architecture to live production.

The role
You own AI systems end to end, from architecture to live production. You will:
Build harnesses: test and eval harnesses for LLM and voice pipelines, simulated conversations, and regression suites that catch quality drops before clients see them
Instrument what ships: tracing of LLM and agent calls (prompts, tool calls, tokens, cost, latency), quality monitoring and regression alerts in production
Design and build microservices: new services and integrations that fit into a platform running in production, including MCP servers and tools for our agents and internal tooling
Optimize AI pipelines: latency, cost, accuracy, and reliability, measured with real metrics
Automate everything repetitive: checks, deployments, agent setup, and routine ops, using scripts, agents, and MCP tooling
Design for safety and privacy: prompt injection, PII handling, call-recording consent, human approval for irreversible actions
Own live production: monitor, debug, and fix real campaigns running real calls

What the role is not
Not a research role. No model training from scratch, no TensorFlow/PyTorch.
Not a prompt-only role. Prompts are one tool among evals, tracing, integrations and ops.
Not a frontend role. React/Next.js is in the stack, but it is not where you will spend your days.

Our stack
TypeScript, NestJS, PostgreSQL, React, Next.js · Retell AI, LiveKit, Deepgram, ElevenLabs, Twilio · Gemini/Vertex AI, OpenAI, Anthropic · Langfuse / OpenTelemetry-style tracing · GCP, Terraform, Docker · Cursor, Claude Code, MCP

Must have
4+ years of engineering experience in any server-side language, with a strong track record of designing, running and supporting microservice architectures in live production
Production experience with LLMs: prompting, structured output, tool calling, RAG, evals
Experience building test or eval harnesses and automating pipelines
Experience with LLM observability: tracing prompts, tool calls, tokens, cost and latency, and using traces to find and fix regressions
AI-native workflow: you do most of your tasks through AI agents and know how to build context for them
Hands-on experience with MCP: you have built or wired MCP servers and tools, not only used them
Strong SQL: comfortable analysing call and campaign data in PostgreSQL
Experience with real-time or webhook-driven systems
You design for safety and privacy: prompt injection, PII, consent, guardrails
English B2+

Nice to have
Voice AI experience (Retell, Vapi, LiveKit, Pipecat, telephony, SIP/WebRTC, streaming audio, STT/TTS, turn-taking and interruptions, latency tuning)
Past experience as a team or tech lead
Forward deployed or solutions engineering experience at a startup
GCP, Terraform, on-prem deployments

Hiring process
Online prescreen
Interview
Decision