Hire Rafi S. - AI Engineer

* AI-Assisted Databricks Medallion Data Mapper (Multi-Agent System) * Designed a canonical silver data model ingesting from multiple bronze sources, applying scientifically grounded data modeling principles to enable reliable LLM reasoning over enterprise schemas. * Built an automated context-engineering pipeline aggregating signals from Unity Catalog, Confluence, and SharePoint into structured prompts for downstream agents. * Orchestrated a multi-agent mapping workflow using Databricks Genie and Claude Opus 4.7 to generate, cross-validate, and arbitrate column mappings — leveraging multiple model perspectives to reduce hallucination. * Validated mappings against human-curated ground truth, achieving 85% accuracy and reducing manual data-engineering effort on bronze-to-silver mapping. * ABAC – Attribute-Based Access Control on Databricks * Led platform-wide migration to Databricks Runtime 16.4+ across workspaces, jobs, and pipelines to unlock Unity Catalog ABAC for PII-sensitive healthcare data. * Designed and implemented row-level filters, column masks, and governed-tag policies replacing legacy RBAC across 50+ data sources. * Drove pilot, UAT, and phased rollout — strengthening HIPAA/PII compliance posture for Optum Financial's regulated workloads.

Setup monitoring for performance, XAI (Explainable Artificial Intelligence), drift, stability, operations and custom metrics for a number of ML and deep learning models in production. Build a generic code template to generate all the common metrics such as lift, R2, RMSE, accuracy, precision and recall. Created script to perform SHAP (SHapley Additive exPlanations) analysis. Extensively used Airflow, GCP dataproc, bigquery, Jenkins and Gitlab to set up pipelines to load data. Created multiple Grafana dashboards with timeseries, histogram and bar charts, and customized layouts. Performed POC to create AI Agents to onboard new models to RT-AIMLOPs platform using VEGAS (Verizon Enterprise GenAI & Agentic Services).

* Project - Personalization AI (PzAI) * ● Built, trained and tested models using PzAI architecture (a novel LLM based recommender system ref paper published by our team) for a number of use cases such as dynamic offers, purchase propensity and channel recommendations * ● Improved model performance by fine tuning models on large datasets using A100/T4 systems on Domino Data Lab platform * ● Wrote complex BigQuery SQLs for daily data ingestion and feeds to model training * ● Performed analytics on large datasets using PySpark to identify key features and extract valuable insights relevant to use cases * ● Implemented dynamic offer recommendation routing via Pega Adaptive Decision Manager (ADM) * ● Build postman testing suite using postman scripts to run 100s of test cases on multiple API endpoints * ● Collaborated with cross-functional teams to RCA a number of defects and followed through to closure * ● Implemented model performance and drift monitoring in prod and retraining strategies. * Project - RT-AIMLOPs * ● Setup monitoring for performance, XAI (Explainable Artificial Intelligence), drift, stability, operations and custom metrics for a number of ML and deep learning models in production * ● Build a generic code template to generate all the common metrics such as lift, R2, RMSE, accuracy, precision and recall * ● Created script to perform SHAP (SHapley Additive exPlanations) analysis * ● Extensively used Airflow, GCP dataproc, bigquery, Jenkins and Gitlab to set up pipelines to load data * ● Created multiple Grafana dashboards with timeseries, histogram and bar charts, and customized layouts * ● Performed POC to create AI Agents to onboard new models to RT-AIMLOPs platform using VEGAS (Verizon Enterprise GenAI & Agentic Services)