AI Engineer

Location: Gurgaon (Hybrid)Experience: 2–5 yearsType: Full-timeAbout NeurogentNeurogent is an AI-native product and engineering partner for the insurance industry — carriers, brokers, MGAs, TPAs and InsurTechs. We're building the AI Operating System for Insurance: reusable agents, accelerators and platforms that take AI from prototype to production.Roughly 70% of enterprise AI pilots never leave PoC. Our entire business is being the team that gets them past that line.The roleAs an AI Engineer at Neurogent, you'll design and build the agentic systems at the core of what we deliver — claims assistance agents, AI SDRs, document extraction pipelines, conversational analytics, real-time scoring services. You'll own these systems from problem framing through to production monitoring.This is an engineering role, not a research role. We care about systems that work reliably on messy enterprise data.What you'll do•⁠ ⁠Design and build production LLM systems: multi-step agents, tool use, RAG pipelines, document understanding, structured extraction•⁠ ⁠Build evaluation harnesses before you build the feature — golden datasets, regression suites, LLM-as-judge where appropriate•⁠ ⁠Take systems to production: latency and cost optimisation, caching, fallback and retry strategies, guardrails, PII handling, audit trails•⁠ ⁠Instrument and monitor deployed AI systems — tracing, drift, quality regression, failure analysis•⁠ ⁠Design human-in-the-loop workflows where full automation isn't safe or acceptable — a core pattern in insurance•⁠ ⁠Partner with data engineers on the retrieval and feature layer, and with full stack engineers on the product surface•⁠ ⁠Contribute to our reusable agent and accelerator library — the IP that makes the next project fasterWhat we're looking for•⁠ ⁠2–5 years in software or ML engineering, with at least 1 year building LLM-based systems that reached production•⁠ ⁠Strong Python; comfortable writing clean, tested, deployable code•⁠ ⁠Deep hands-on experience with LLM APIs and orchestration — prompting, function/tool calling, structured outputs, context management, agentic loops•⁠ ⁠Practical RAG experience: chunking strategies, embeddings, vector stores, hybrid and re-ranked retrieval, and honest evaluation of it•⁠ ⁠Experience measuring AI system quality — you can explain how you knew your system was actually good•⁠ ⁠Cloud and MLOps fundamentals: containers, CI/CD, model/prompt versioning, monitoring•⁠ ⁠Clear communication — you can explain a model's failure mode to a business stakeholder