AI Engineer

Key Responsibilities

• Translate CX and business needs into AI solution designs spanning NLU/LLM, RAG, ASR/TTS and back-end integrations.
• Define functional and non-functional requirements: latency, resiliency, observability, security, cost-per-conversation.

• Design and build agentic AI solutions — single-agent, multi-agent, tool-using and orchestrated patterns — using modern frameworks and frontier models.
• Build and optimise NLU/LLM components: classification, NER, summarisation, RAG over enterprise knowledge bases.
• Engineer prompts, tool schemas, context windows and token budgets for accuracy, latency and cost.
• Develop data and knowledge pipelines for ingestion, cleansing, PII redaction and evaluation datasets.

• Build evaluation harnesses (offline eval sets, LLM-as-judge, regression suites, A/B testing) and treat eval as a first-class deliverable, not an afterthought.
• Drive continuous improvement via conversation analytics, error triage and failure-mode analysis.

• Implement CI/CD for models and prompts, with feature flags, canary releases and rollback.
• Track experiments, lineage, and model/prompt versions.
• Operate services in production against SLOs with monitoring, tracing, alerting and incident response.

• Apply practical guardrails for safety, bias, hallucination, prompt-injection and jailbreak resistance.
• Enforce GDPR / LOPDGDD: consent, minimisation, retention, access control, auditability.

• Work alongside conversation designers, software engineers, data scientists, QA and Ops.
• Produce clear design docs, runbooks and stakeholder updates.

Skills Knowledge and Expertise

• Hands-on experience delivering conversational AI — voicebots, chatbots or virtual assistants — including conversational flows, open-ended interactions, NLU and generative AI.
• Production experience building agents with generative AI: agentic architectures (single-agent, multi-agent, tool/function-calling, planner–executor patterns), RAG, and orchestration.
• Working knowledge of frontier LLM providers and platforms — e.g. Anthropic Claude, OpenAI, Google Gemini, Amazon Bedrock, Microsoft Azure AI — and a practical sense of when to use which.
• Strong prompt engineering and context engineering skills across multiple LLM families, including system prompt design, structured output, and grounding strategies.
• An evaluation mindset: you measure agent quality with eval sets, traces and metrics — not vibes.
• Awareness of AI ethics, bias and safety, with practical experience mitigating them in deployed systems.
• Token, latency and cost optimisation — model routing, caching (incl. prompt caching), context compression, retrieval tuning — as a core engineering discipline.
• Comfort with agentic development workflows — using AI coding assistants and AI co-work / pair-development models (Claude Code, Copilot, Cursor or equivalent) as part of your day-to-day delivery.
• Solid software engineering fundamentals: Python and/or JavaScript/Node, Git, IDEs, Agile delivery.
• Customer-facing skills: requirements gathering, design, validation and stakeholder communication.
• Working proficiency in Spanish and English.

• Cloud services, primarily AWS, with containerised deployment via Docker / Kubernetes.
• Contact centre integrations (e.g. Genesys, Avaya, Cisco, Amazon Connect) and Meta / WhatsApp Business / digital channel integrations.
• Experience with ASR/TTS and voice-specific concerns: barge-in, turn-taking, ASR error recovery, latency budgets.
• Speech/NLU recognition models and continuous-improvement loops (Microsoft STT, MS CLU, Google, Amazon).
• Observability for LLM systems: tracing, eval pipelines, online quality monitoring.
• Familiarity with emerging agent interoperability standards (e.g. MCP, A2A) and human-in-the-loop / escalation patterns.
• Microsoft Bot Framework, Genesys Dialog Engine or VXML exposure.
• Consulting experience analysing existing bot estates, evaluating KPIs (containment, CSAT, AHT, transfer rate) and recommending improvements.

• Exposure to Google Conversational Agent / Playbooks or DialogFlow — useful if you've worked with it, but not required.

Benefits