Data Scientist | Applied AI • NLP • Model Evaluation • RAG

I build applied data science and AI projects focused on NLP, model evaluation, retrieval systems, and measurable decision-support workflows using Python, SQL, scikit-learn, PyTorch, Transformers, PostgreSQL/pgvector, and FastAPI.

I'm Rihua Van Steenburgh, a Data Scientist focused on applied AI, NLP, model evaluation, and retrieval systems.

My portfolio emphasizes leakage-safe validation, model comparison, temporal evaluation, error and failure analysis, human-in-the-loop decision making, retrieval quality, and grounded AI applications.

I'm currently pursuing a Master of Science in Information Technology with a Data Analytics concentration at Middle Georgia State University, expected in December 2026.

I also bring hands-on cloud data engineering experience with Azure, AWS, Databricks, PySpark, dbt, and data pipelines, giving me a strong data foundation for building and evaluating AI systems.

• Applied Data Science & NLP: scikit-learn, TF-IDF, Linear SVM, PyTorch, Transformers, DistilBERT, text classification, and leakage-safe evaluation

• Model Evaluation: model comparison, temporal validation, group-aware validation, error analysis, failure analysis, routing metrics, and human-in-the-loop decision policies

• Retrieval & Applied AI: hybrid retrieval, PostgreSQL/pgvector, full-text search, RRF, grounded generation, citation validation, abstention handling, and retrieval evaluation

• Data & Cloud Foundations: Azure, AWS, Databricks, PySpark, dbt, SQL, data pipelines, data quality, Docker, and GitHub Actions

• Programming & Analysis: Python, SQL, pandas, NumPy, Statistical Analysis

• Machine Learning & NLP: scikit-learn, TF-IDF, Linear SVM, PyTorch, Transformers, DistilBERT, Text Classification

• Model Evaluation: Macro F1, Accuracy, Model Comparison, Error Analysis, Leakage-Safe Validation, Group-Aware Validation, Temporal Validation

• Applied AI & RAG: FastAPI, PostgreSQL/pgvector, Embeddings, Vector Search, Full-Text Search, Reciprocal Rank Fusion (RRF), Retrieval Evaluation, Grounded Generation, Citation Validation

• Data & Cloud: Azure Data Factory, ADLS Gen2, Databricks, PySpark, Delta Lake, dbt, AWS, Redshift Serverless

• Engineering & Tools: Docker, Git/GitHub, GitHub Actions, Airflow, Jupyter Notebook, Power BI

Leakage-safe CFPB complaint-classification and selective-routing study comparing TF-IDF + Linear SVM with a frozen DistilBERT challenger.

• Identified 39.39% normalized-text leakage in the original test split and redesigned evaluation to achieve zero development/test overlap
• Compared classical NLP and transformer approaches using classification, coverage, routed accuracy, and misroute metrics
• Added human-review routing and retrospective temporal evaluation
• Retained V1 as the temporally validated benchmark while documenting V2 trade-offs

Tech: Python, scikit-learn, TF-IDF, Linear SVM, PyTorch,Transformers, DistilBERT, model evaluation

Hosted, non-production hybrid RAG application for grounded NYC 311 documentation Q&A and bounded analytics.

• Combines semantic retrieval with PostgreSQL full-text search using deterministic Reciprocal Rank Fusion
• Uses PostgreSQL/pgvector, FastAPI, validated citations, explicit abstention handling, and bounded analytics
• Hybrid retrieval reached 83.9% Recall@5 and 92.9% expected-source retrieval on the approved local evaluation
• Includes retrieval experiments, failure analysis, and a hosted Vercel → Render → Neon application path

Tech: Python, PostgreSQL, pgvector, embeddings, vector search,RAG, Streamlit, Docker, pytest, GitHub Actions

Azure lakehouse pipeline using Azure Data Factory, ADLS Gen2, Databricks, PySpark, SQL, and Delta Lake to produce validated Bronze, Silver, Gold, fact, dimension, and analytics-mart outputs.

AWS batch data pipeline using Docker, ECS/Fargate, EventBridge, S3, Redshift Serverless, SQL, and dbt with data-quality tests, CI checks, runbooks, and cloud execution proof.

• 💼 LinkedIn
• 🌐 Portfolio
• 🧑‍💻 GitHub

Thanks for stopping by! ✨