AI Engineering

Role Purpose

1-AI Model Development & Production Deployment

• Build, train, evaluate, and deploy production-grade AI/ML models on GCP Vertex AI - covering supervised, unsupervised, and generative AI approaches as required by the use case
• Deploy Wave 1 priority models to GCP production inference endpoints: FWA Service Overutilization, FWA Duplicated Claims, FWA Provider Network Collusion, Document OCR Extraction, Document Classification, and Service Code Mapping
• Implement automated model retraining pipelines using Vertex AI Pipelines - no model should require manual retraining once in production
• Evaluate and implement Gemini (approved by Bupa Arabia) for OCR and FWA detection - assess accuracy, latency, and cost; deploy on GCP

2-MLOps & Model Lifecycle Management

• Build and maintain MLOps infrastructure: Vertex AI Pipelines for training/evaluation, Vertex AI Model Registry for versioning, and Vertex AI Model Monitoring for drift detection
• Implement CI/CD for model deployment - models promoted from development to staging to production via automated pipelines with quality gates; no manual uploads
• Design and operate feature engineering pipelines: build and maintain the Vertex AI Feature Store from Gold layer BigQuery data - features versioned, documented, and reusable across use cases
• Maintain model cards for every production model - documenting training data, evaluation metrics, known limitations, drift thresholds, and business use context

3-Cross-Functional AI Solution Delivery (FWA & Claims)

• Work directly with the FWA team to translate fraud patterns, clinical rules, and investigator knowledge into model features, labels, and evaluation criteria for FWA detection models
• Deliver AI solutions for claims intelligence: duplicate claims detection, service code mapping accuracy, and provider network anomaly detection - all with explainable outputs usable by claims adjusters
• Build pre-authorization scoring models integrated with the operational Pre-Auth workflow - providing real-time risk scores to clinical teams within SLA (\<2 seconds inference latency)
• Collaborate with actuarial teams on loss ratio prediction and IBNR estimation models - ensuring all models use the 3+ year historical dataset loaded by the platform team

4-Document Intelligence & OCR Engineering

• Deploy and optimize the Document OCR Extraction model to GCP Vertex AI production - 7 months of development across 12+ open-source models completed; this role deploys it with serving infrastructure and output post-processing pipeline
• Build Document Classification pipeline: classify incoming insurance documents by type with high accuracy using Vertex AI
• Evaluate Gemini's multimodal capabilities for document understanding - compare against existing OCR models on accuracy, cost per document, and latency for Bupa Arabia's document types
• Integrate document intelligence outputs with downstream business workflows: pre-auth queue management, claims processing, and member enrollment - structured data flowing into BigQuery Gold layer

5-Generative AI & Gemini Enablement

• Lead evaluation and implementation of Gemini (approved for Bupa Arabia) across priority use cases - OCR, document classification, FWA detection, and member service intelligence
• Design responsible AI guardrails for Gemini deployments - input validation, output filtering, hallucination detection, and audit logging for all generative AI interactions with business data
• Build internal AI capability: run knowledge-sharing sessions with data engineers and analysts on Gemini APIs, Vertex AI generative AI features, and prompt engineering best practices
• Stay current with GCP AI/ML releases - evaluate new Vertex AI capabilities and Gemini model versions for applicability to Bupa Arabia use cases

6-AI Data Quality, Lineage & Ethics

• Ensure all training datasets are sourced from the Gold layer - no model trained on uncleaned, ungoverned data from the legacy EDWH or raw SSIS extracts
• Document and maintain full data lineage for every model: training data source, feature engineering logic, evaluation dataset, and production data feed - captured in GCP Dataplex
• Apply Bupa Arabia's data classification standards to all model inputs and outputs - PII, PHI, and sensitive clinical data handled in compliance with PDPL and NDMO requirements
• Conduct bias audits on FWA and clinical models - ensure models do not exhibit demographic bias and that outputs are explainable to clinical and compliance stakeholders

Skills

• Vertex AI - Pipelines, Model Registry, Feature Store, Model Monitoring, online/batch inference
• Python - ML engineering (scikit-learn, XGBoost, TensorFlow, PyTorch), data processing, pipeline scripting
• GCP - BigQuery ML, BigQuery data access for training data, feature engineering queries
• MLOps - CI/CD for models, automated retraining, drift detection, model versioning
• Gemini API / Vertex AI Generative AI - prompt engineering, multimodal models, RAG patterns
• Document Intelligence / OCR - Vertex AI Document AI, Tesseract, LayoutParser, multimodal LLMs
• GCP - Pub/Sub, Dataflow, Cloud Functions for real-time AI inference pipelines
• NLP / Text Classification - text preprocessing, embedding models, classification fine-tuning
• Fraud / Anomaly Detection - unsupervised methods, graph analytics, rule-ensemble models
• Explainability / Responsible AI - SHAP, LIME, bias auditing, model cards
• SQL / BigQuery - feature engineering queries and training data preparation
• Git + CI/CD - model deployment pipelines and experiment tracking (MLflow or Vertex Experiments)

Education