Applied AI Engineer - Florida Blue - Jacksonville, FL
(2021-09)
Partnered directly with the Chief Technology Innovation Officer and business executives to identify high-value internal automation opportunities, scoping 8 AI initiatives from discovery through prototype to production within 18 months, each measured against real user adoption and time-savings KPIs.
- Architected and shipped a six-agent production platform on Vertex AI Agent Builder using Gemini Pro and Claude claude-sonnet-4-6 as core reasoning models, delivering a full Orchestrator to Supervisor to Worker agent hierarchy; designed MCP (Model Context Protocol) server contracts defining hard-typed input/output schemas, shared context persistence, rate limiting policies, and guardrails to prevent agent state divergence across independently deployed services.
- Built end-to-end AI applications covering the full stack—React and Streamlit user interfaces, FastAPI application services, LLM model integration via Vertex AI and Anthropic APIs, Vertex AI RAG Engine knowledge bases, and Cloud Run/GKE deployment—enabling employees to surface information, automate repetitive processes, and complete real work through AI-powered internal tools.
- Designed and deployed a rapid prototyping workflow using Claude Code and Cursor to go from rough executive idea to interactive demo in under 5 business days; prototypes were tested with real users, feedback was captured, and solutions were either hardened to production or deliberately abandoned based on measured outcomes.
- Implemented a LangGraph-based agent orchestration backbone modeling workflows as typed state machines with explicit retry policies, timeout budgets, compensating rollback operations, and run-context persistence via Cloud Firestore; integrated tool calling, retrieval-augmented generation, and structured reasoning loops following full-stack AI patterns aligned with enterprise guardrails.
- Built an automated learning cycle integrating auto-classification via Gemini embeddings, auto-validation against ground-truth references, sampled human review queue routing, and governed writeback to Vertex AI RAG Engine knowledge bases, reducing manual review time from approximately 9 hours per day to under 2 hours.
- Developed an adversarial LLM-as-judge scoring loop using Gemini as the evaluator model, scoring agent outputs against validation schemas with non-convergence fallback re-generation and escalation queuing to human SME reviewers; established nightly Vertex AI Evaluation Harness regression pipelines publishing pass rates, latency distributions, and cost-per-task metrics to Cloud Monitoring dashboards with automated PagerDuty alerting.
- Developed a multi-region retrieval system using Vertex AI RAG Engine with regional knowledge base overlays enabling region-specific context injection into agent prompts and supporting simultaneous retrieval from geographically partitioned knowledge stores.
- Moved successful prototypes into reliable production solutions with appropriate security including VNet/Private Endpoints, Managed Identities, RBAC, and GCP IAM zero-trust policies, along with pytest integration test suites, type-checked Pydantic codebases, monitoring, and human oversight workflows.
- Collaborated with business stakeholders and domain SMEs to translate workflow requirements into agent tool specifications, defining tool input contracts, expected output formats, and validation acceptance criteria for each agent in the multi-agent system.
- Mentored a team of three engineers on Vertex AI Agent Builder patterns, MCP server design, and LangGraph state machine modeling, conducting weekly code reviews and authoring internal engineering playbooks covering agent observability, prompt versioning, and schema governance.
ML Engineer - EverBank - Jacksonville, FL
(2019-07 - 2021-09)
Worked directly with operations and compliance business users to discover high-friction manual processes; translated pain points into ML solution designs, shipped working prototypes within 2-week sprints, and iterated based on structured user feedback before committing to production buildout.
- Designed and deployed a GCP AI Platform AutoML tabular classification system trained on historical financial data quality annotations, integrating the production scoring endpoint into daily data ingestion workflows and reducing manual data stewardship work from approximately 6 hours per day to under 90 minutes.
- Built multi-step financial data validation pipelines in Python orchestrated via Apache Airflow, incorporating rule-based anomaly detection, regulatory threshold checking, and automated exception report generation routed to compliance reviewers via structured alerting workflows.
- Implemented a BERT-based text classification model fine-tuned on internal financial compliance documents to auto-tag incoming regulatory filings by category and risk level, enabling faster triage by the compliance team and reducing manual document review time by approximately 60%.
- Engineered a real-time payment event processing pipeline consuming Azure Event Hub streams, routing events through a scikit-learn fraud signal scoring model deployed on Azure ML, and publishing structured risk scores to downstream fraud detection systems with end-to-end latency under 90 seconds.
- Built a keyword and TF-IDF based document retrieval system over a financial regulatory knowledge base using Elasticsearch, enabling compliance analysts to surface regulation-specific validation rules and policy guidance without manual document search.
- Led cloud migration re-architecting 20+ legacy ETL processes onto GCP and Azure, redesigning workflows with GCP AI Platform Pipeline components for ML feature engineering alongside Apache Spark and Azure Databricks batch processing, cutting average end-to-end processing time from over 6 hours to under 3 hours.
- Configured model evaluation pipelines benchmarking AutoML and scikit-learn classification model accuracy against human-labeled financial datasets, establishing F1-score and precision/recall baselines and implementing automated re-evaluation on every model version promoted to production.
- Developed Python production services with full type-checking using mypy and Pydantic, pytest test suites covering unit, integration, and property-based tests, and Azure DevOps CI pipelines enforcing linting, type checks, and coverage gates before any production deployment.
Data Engineer - Wachter - New Jersey, NJ
(2018-05 - 2019-07)
Built and maintained the internal network operations analytics data platform, designing Azure Data Factory pipelines to ingest structured event logs from Wachter's nationwide IoT sensor network, structured cabling systems, and network monitoring tools into ADLS Gen2 for operational reporting and anomaly detection.
- Designed and trained a scikit-learn gradient boosting classification model on labeled network infrastructure event logs to auto-categorize anomaly types across client telecom deployments, integrating the model into an operational dashboard used daily by the operations team and reducing incident triage time from approximately 40 minutes to under 4 minutes.
- Built multi-step batch scoring workflows using Apache Airflow
