Generative AI Engineer

ProjectBookMe AI is an AI receptionist that works inside a business's existing WhatsApp number, handling bookings, rescheduling and cancellations in conversation, without a human operator. It is built for small service businesses — barbershops, salons, auto service, clinics, restaurants — where a missed message is a lost customer, and it is currently moving from working product into the hands of the first paying businesses. The next phase is a research and development programme extending the agent onto new channels and into live voice, and building the scheduling engine underneath it.Profile3+ years, with LLM systems in production rather than prototypes.2–3 delivered projects, at least one of them an LLM or ML system in real use. Work done as an employee or contractor counts, including internal systems built for a company rather than sold externally.Higher education degree completed.Minimum 6 months working daily in an agentic IDE — Cursor, Claude Code or equivalent.ResponsibilitiesRealtime voice agent. Take the agent from text to live phone conversation, with the latency and turn-taking behaviour a real call demands.Speech understanding. Make spoken input work as reliably as written input, including noisy and unstructured audio.Knowledge-grounded answering. Build retrieval over business-provided knowledge so the agent advises accurately and declines cleanly when it doesn't know.Agent configurability. Make agent behaviour adjustable per business, and build the assistant the owner interacts with directly.Production agent engineering. Own reliability, observability and cost of the agent in production, plus the predictive and classification work around it.TechnologiesRequiredPython and TypeScript (working, or fast ramp — the agent runs inside the Next.js app)LLM agents in production: tool/function calling, state management, structured outputPrompt engineering and prompt versioning as an engineering disciplineRetrieval over a document corpus, with grounded answers and clean refusal when the answer isn't thereLangchain, openai, geminiVoiceGemini Live, OpenAI Realtime, ElevenLabs — the direction we're starting fromA candidate who has used something else and can argue why it's better is exactly what we want here; this stack is a starting point, not a constraintStack in useLangChain / LangGraph, Gemini and OpenAI, Meta WhatsApp Cloud APIPostgres/Prisma, Next.js, VercelNice to havea vector store in productionLLM tracing and observability toolingML experience : scikit-learn