Senior AI/ML Engineer

Job Responsibilities

Systems Thinking & Solution Design

• Decompose business problems into technical components — identify what needs an ML model, what needs a rule, what needs a knowledge graph, and what needs a human in the loop.
• Design end-to-end system architectures for AI/ML solutions: data ingestion, feature engineering, model training, serving, monitoring, and feedback loops — with clear rationale for each design choice.
• Anticipate failure modes, bottlenecks, and degradation paths; design for observability and graceful failure from day one, not as an afterthought.

Machine Learning Engineering

• Build, train, evaluate, and tune ML models for insurance use cases such as claims severity prediction, fraud detection, risk scoring, document classification, and policy matching.
• Implement feature engineering pipelines that are reproducible, versioned, and decoupled from model training — so features can be reused across models and teams.
• Go beyond accuracy: evaluate models on business-relevant metrics (cost-of-error, false-positive impact, fairness), not just AUC and F1.

Generative AI & LLM Engineering

• Build production-grade GenAI applications using LLMs, Retrieval-Augmented Generation (RAG), tool/function calling, and Agentic AI frameworks (LangGraph, CrewAI, AutoGen).
• Design prompt and context engineering strategies grounded in domain-specific knowledge — insurance terminology, regulatory language, claims narratives.
• Build evaluation pipelines for GenAI outputs: correctness, faithfulness, relevance, latency, and cost — using frameworks such as RAGAS, LangSmith, or custom harnesses.

Knowledge Engineering & Enterprise Intelligence

• Ability to work with model domain knowledge as structured, queryable assets — ontologies, knowledge graphs, taxonomy mappings, and enriched metadata layers — that ground AI systems in the customer’s business vocabulary.
• Build enterprise intelligence capabilities that combine structured data, unstructured documents, and domain rules into unified, AI-ready knowledge layers for downstream consumption by ML models and GenAI applications.
• Collaborate with domain experts to codify business rules, classification hierarchies (e.g., ISO class codes, NAICS mappings, loss-cause taxonomies), and decision logic into reusable knowledge assets.

Cloud-Native AI at Scale

Deploy and operationalize AI/ML workloads on cloud-native platforms with infrastructure that scales elastically and costs predictably.

Reference architecture on Databricks: ingest raw data via Auto Loader into Bronze Delta tables → cleanse and enrich through Silver-layer notebooks orchestrated by Databricks Workflows → compute features using Feature Store → train and log models in MLflow with experiment tracking → register production-ready models in Unity Catalog’s Model Registry → serve via Model Serving endpoints behind REST APIs → monitor for drift using Lakehouse Monitoring → trigger automated retraining when quality thresholds breach. This end-to-end loop — from landing zone to live inference to feedback — is the kind of system you will own.

• Containerize models and applications using Docker; orchestrate with Kubernetes where appropriate; serve via REST APIs or event-driven architectures.
• Build and maintain MLOps/LLMOps pipelines for automated training, evaluation, deployment, monitoring, versioning, and retraining — using tools such as MLflow, Unity Catalog, Feature Store, Model Registry, and CI/CD automation (GitHub Actions, Azure DevOps, Databricks Asset Bundles).
• Implement model and data observability: drift detection, data quality checks, latency/throughput monitoring, and alerting — so the team knows when something degrades before the customer does.

Data Pipelines for AI

• Design and build data ingestion pipelines and ETL/ELT workflows using Apache Spark, Databricks, and Delta Lake.
• Build Lakehouse-based platforms; implement data quality, validation, and lineage frameworks that AI workloads depend on.
• Work with messy, heterogeneous, real-world data — inconsistent schemas, missing values, multi-format source files — and make it model-ready without losing traceability.

Business Problem Orientation

• Participate in requirement discussions and discovery sessions with business stakeholders and clients; understand the problem before reaching for a model.
• Translate business outcomes (reduce claims leakage, accelerate underwriting, improve triage accuracy…