Conversation AI Engineer

PURPOSE OF THE ROLE You build customer facing AI agents that answer real enquiries correctly, in English and Arabic, and you are accountable for how they behave in production. Our first deployment for this role is a bilingual customer service agent for a major client in the UAE energy sector, covering web, messaging and voice. WHAT YOU WILL DO • Design and build AI agents that resolve customer enquiries from approved knowledge sources, with answers traceable to source content • Own retrieval quality: how content is prepared, chunked, indexed and refreshed, and how the agent behaves when it has no good answer • Build and tune bilingual behaviour, including language detection, switching mid conversation, and Arabic responses that read naturally to a Gulf audience • Define escalation rules and the handover to human agents, so the customer is never left in a loop • Set up evaluation before launch: test sets, accuracy measurement, regression checks on every change • Instrument the agent so resolution rate, escalation volume and response accuracy are visible to the client in reporting • Work in the client's environment under their security and data residency requirements • Hand over documentation and run training so the client's own team can operate and extend the agent WHAT SUCCESS LOOKS LIKE IN THE FIRST 90 DAYS • A working agent in a client environment, answering a defined enquiry set in both languages • An evaluation set the client agrees represents their customers, with measured accuracy against it • Escalation and handover rules signed off by the client's service team ESSENTIAL EXPERIENCE • Three years or more building software, with at least one year on LLM based applications that reached real users • Production experience with retrieval over enterprise content, and a clear view on why retrieval quality fails • Strong Python, and comfort working inside someone else's codebase and cloud account • Experience designing conversation flows, including failure paths, not only the ideal path • Evidence of measuring an AI system rather than demonstrating it • Able to work onsite with client teams and explain technical decisions to non technical stakeholders DESIRABLE • Arabic language skills, particularly Gulf dialect handling in speech or text • Experience with contact centre platforms, WhatsApp Business or voice channels • Work in a regulated environment: banking, energy, healthcare or government • Exposure to guardrails, prompt injection defence and content safety in customer facing systems ESSENTIAL EXPERIENCE u2022 Three years or more building software, with at least one year on LLM based applications that reached real users u2022 Production experience with retrieval over enterprise content, and a clear view on why retrieval quality fails u2022 Strong Python, and comfort working inside someone else's codebase and cloud account u2022 Experience designing conversation flows, including failure paths, not only the ideal path u2022 Evidence of measuring an AI system rather than demonstrating it u2022 Able to work onsite with client teams and explain technical decisions to non technical stakeholders DESIRABLE u2022 Arabic language skills, particularly Gulf dialect handling in speech or text u2022 Experience with contact centre platforms, WhatsApp Business or voice channels u2022 Work in a regulated environment: banking, energy, healthcare or government u2022 Exposure to guardrails, prompt injection defence and content safety in customer facing systems