Applied AI Engineer and Data Engineer building production RAG, document-intelligence, evaluation, and data-platform systems with Python, SQL, AWS, Databricks, and Kubernetes.

LinkedIn · AWS Machine Learning Engineer – Associate · Databricks Data Engineer Professional

My progression is deliberate: four years in production Data Engineering, graduate work in data science and AI, and current ownership of production Applied AI systems. I focus on the full engineering loop—understand the workflow, choose the simplest defensible approach, measure quality, build for failure, and make the result usable by the people operating it.

• Own two production AI workstreams across requirements, architecture, implementation, deployment, monitoring, incident response, and stakeholder handoff.
• Improved document-extraction F1 from 0.84 to 0.92 across 1,200 labeled documents, with 8% of low-confidence outputs routed for human review.
• Increased retrieval Recall@5 from 0.71 to 0.88 on a 250-question evaluation set and reduced unsupported answers from 14% to 4%.
• Design golden datasets, failure taxonomies, acceptance thresholds, A/B evaluations, regression gates, and release criteria.
• Choose among deterministic logic, classical ML, retrieval, LLMs, and human review according to measured task behavior.
• Operate FastAPI services on Kubernetes with validation, timeouts, retries, idempotency, fallbacks, observability, audit logging, and controlled rollback.

• Assisted students and professors with assignments, examinations, SQL, data structures, algorithms, analytics, and technical coursework.
• Supported faculty research and worked with the Program Director on the design of a new Artificial Intelligence and Natural Language Processing course.

• Operated 24 production pipelines processing 50 GB daily with 99.5% on-time completion across four business accounts.
• Built batch, incremental, and streaming workflows with Python, SQL, PySpark, Airflow, Kafka, Databricks, Delta Lake, and Unity Catalog.
• Reduced pipeline runtime by 38% and monthly compute cost by 24% through incremental processing, partition optimization, and Spark tuning.
• Automated eight recurring reporting workflows, saving approximately 32 hours per full reporting cycle, and reduced new-workflow setup from three days to under four hours.
• Implemented reconciliation, schema-drift controls, retries, idempotency, backfills, monitoring, runbooks, incident recovery, and controlled production releases.

Python, FastAPI, Pydantic, REST APIs, RAG, document intelligence, structured outputs, XGBoost, scikit-learn, deterministic validation, human-review routing, evaluation datasets, F1, Recall@k, unsupported-answer analysis, A/B evaluation, regression gates, Ollama and Llama local inference, FAISS and Chroma comparisons, LangSmith traces, Docker, GitHub Actions, Terraform, Kubernetes service operation, observability, rollback, incident recovery, and AI security controls.

Python, SQL, Pandas, Apache Airflow, Apache Kafka, Databricks, PySpark, Spark SQL, Delta Lake, Databricks Workflows, Unity Catalog, batch and incremental ingestion, streaming, backfills, idempotency, data quality, reconciliation, schema-drift handling, dbt, Snowflake, BigQuery, PostgreSQL, SQL Server, AWS S3, Lambda, IAM, CloudWatch, Power BI, Excel, Streamlit, CI/CD, performance tuning, and cost optimization.

Timeouts, retries, idempotency, fallbacks, rollback procedures, incident recovery, structured logging, audit trails, prompt-injection testing, output validation, least-privilege tools, sensitive-data filtering, approval gates, rate limits, configuration versioning, requirements discovery, architecture review, code review, testing, UAT, mentoring, operational handoff, runbooks, and reusable engineering standards.

Agent orchestration, policy-as-code, LangGraph-compatible workflows, Qdrant, Sentence Transformers, NVIDIA NIM, Groq, TypeScript, React, Supabase, PostgreSQL row-level security, MLflow concepts, Docker Compose, Helm, Kubernetes manifests, Prometheus, Grafana, Sentry, tamper-evident audit patterns, tool-permission boundaries, and healthcare workflow prototypes.

Machine learning, natural language processing, analytics, visualization, databases, big-data systems, algorithms, data structures, research support, and AI and NLP course design.

• Built a reproducible evaluation workflow using a 25-question golden set, three quality dimensions, and ten failure categories to compare retrieval and answer behavior.
• Produced reviewable evidence for retrieval quality, groundedness, failure analysis, and release decisions.
• Tools: Python, FastAPI, Pydantic, BM25, vector retrieval, JSON, and automated tests

• Built a private local governance project that routes six agent-action outcomes through deterministic policy checks, evidence requirements, and explicit approval boundaries.
• Added restricted tools, least-privilege permissions, sensitive-data filtering, audit events, rate controls, and human approval for higher-risk actions.
• Tools: Python, FastAPI, Pydantic, policy-as-code, structured logging, and automated tests

• Built a local-first document workflow with typed extraction, deterministic validation, review queues, and testable failure handling.
• Kept model output separate from approval decisions so uncertain results remain visible and recoverable.
• Tools: Python, FastAPI, Pydantic, Ollama, Docker, GitHub Actions, and automated tests

Additional public evidence: RestartOS, a simulated manufacturing-recovery prototype, and PharmaTrack, an AI-assisted personal full-stack prototype with deterministic business rules and audit-oriented controls.

• AWS Certified Machine Learning Engineer – Associate · issued July 2026 · expires July 2029
• Databricks Certified Data Engineer Professional · issued July 2026 · expires July 2028

Master of Science in Data Science and Analytics · New England College · January 2024–June 2025 · GPA 3.96

Employer systems and metrics describe professional work. Linked repositories are personal portfolio projects, simulations, learning demonstrations, or design concepts; they do not contain employer code or confidential data and are not presented as commercial deployments unless a repository provides explicit public evidence.