Senior AI/ML Engineer

Senior AI/ML Engineer (Boring Host AI)About Boring HostBoring Host AI is the intelligent operating layer above a property manager’s PMS. We unify guest conversations across channels (OTA inboxes, SMS, WhatsApp, email, and voice), ground answers in PMS truth, and automate the repetitive 80% with guardrails so teams scale without losing the “personal hospitality at scale” feel.We are building AI that behaves like a great ops teammate: fast, consistent, correct, and humble when uncertain.The roleWe’re hiring a Senior AI Engineer to own core intelligence systems end to end: retrieval and knowledge, agent behavior, confidence gating, evaluations, and production reliability. You will work on real-world problems where “mostly works” is not good enough.You will ship models and systems that directly touch guests, revenue, and reviews.What you’ll buildA high-accuracy, PMS-grounded response engine that can answer guest questions with citations and context from listings, reservations, policies, and house rulesA confidence and risk gating system (autopilot only when safe) with fallback behaviors: draft, ask clarifying questions, or escalate to humansA self-improving knowledge layer (RAG plus structured data) that learns from unknown inquiries, reduces repeated escalations, and avoids hallucinationsProduction-grade LLM agent tooling: function calling, tool orchestration, conversation memory, and safe state management across channelsEvaluation harnesses that measure what matters: correctness, policy compliance, tone, latency, cost, and escalation rateThe “ops brain” behind revenue automation: upsell eligibility checks (availability, rules), offer generation, payment workflows, and task triggeringVoice intelligence foundations (if relevant): call summarization, intent extraction, action dispatching, and handoff logicResponsibilitiesOwn AI features from prototype to production: design, implementation, testing, monitoring, iterationBuild and maintain retrieval pipelines and knowledge stores (structured + unstructured), including chunking, indexing, reranking, and freshness handlingCreate robust evals and red-team suites using real operational edge cases (late check-in, lockouts, refunds, maintenance triage, policy disputes)Optimize for reliability and cost: caching, prompt and context efficiency, latency targets, and graceful degradationPartner with product and ops to translate messy hospitality reality into crisp AI behaviors and guardrailsEstablish best practices for observability: traces, error taxonomies, confidence metrics, and feedback loops from human overridesMust-have qualifications3+ years building production ML/AI systems, including at least 2 years working with modern LLM stacks in real productsStrong Python skills and experience shipping services (APIs, workers, event-driven systems)Deep understanding of RAG systems and information retrieval tradeoffs (precision vs recall, freshness, grounding, reranking)Practical experience with evaluation and testing for LLMs (offline evals, online metrics, regression suites, A/B testing)Proven track record of building reliable systems under ambiguity, with strong debugging instincts and ownership mindsetNice-to-have qualificationsExperience with multi-channel messaging platforms, customer support automation, or hospitality / marketplace opsExperience with voice agents or telephony workflows (intent, summarization, escalation)Familiarity with vector databases, hybrid search, and knowledge graph conceptsExperience with TypeScript/Node in addition to Python (helpful for integrations)Experience with privacy, compliance, and secure handling of customer data in SaaS systemsWhat success looks like (first 90 days)You ship at least one major improvement to autopilot safety (confidence gating + eval coverage)Hallucination-driven escalations drop measurably, and “wrong but confident” responses become rareWe gain a repeatable eval and release process that prevents regressions as we ship fasterThe AI system becomes more grounded in PMS truth, with clearer failure modes and better handoffsWhy this is interestingHospitality ops is a perfect stress test for applied AI: multi-channel, time-sensitive, policy-heavy, emotionally charged, and full of edge cases. If you like building AI that actually works in the wild, this is your playground.How to applyA short note on what you’ve shipped in production (links if possible)A resume or LinkedInOne example of a hard reliability problem you solved (briefly: what broke, how you diagnosed it, how you fixed it)