Azure · Databricks · AI / ML

Data systems that react.
Platforms that endure.

I design reliable cloud data platforms and applied AI systems—from streaming ingestion and governed lakehouses to machine learning, RAG, model deployment, and production-ready agents.

Explore projects

68tables in a migration program

14intraday orchestration windows

34hands-on AI / ML notebooks

2public engineering repositories

Featured engineering

Proof, not a skills list.

Two public repositories show both depth and breadth: a production-style Databricks LLMOps system and a hands-on portfolio spanning classical ML, deep learning, computer vision, NLP, RAG, and deployment.

Public GitHub repository

Arxiv Curator

An AI research assistant that ingests the latest arXiv papers, builds governed retrieval, serves a stateful agent, and continuously evaluates production traces. It uses the Databricks platform end to end rather than stopping at a notebook demo.

DatabricksVector SearchGenieLakebaseMLflowUnity CatalogAsset Bundles

View repository GitHub profile

System pathmain / monitored

IngestarXiv API · PDFs · Delta tables

RetrieveParsed chunks · Vector Search

ReasonMCP tools · Genie · memory

ServeUC model · AI Gateway · endpoint

EvaluateMLflow traces · judges · dashboard

AI-ML-Projects · Public GitHub repository

Applied AI & Machine Learning Portfolio

A completed body of notebook-based work that follows the full ML lifecycle: problem framing, data preparation, exploration, feature engineering, model training, evaluation, interpretation, and application deployment.

34notebooks

8skill areas

68tracked files

ML / 01

Medical Knowledge RAG

Builds document ingestion, chunking, embeddings, vector retrieval, prompt engineering, and grounded question answering over a medical reference corpus.

LangChain · Chroma · GTE embeddings · Llama.cpp ML / 02

EasyVisa Approval Prediction

Explores 25K+ cases, treats class imbalance, compares ensemble classifiers, tunes models, and evaluates the business tradeoff between recall and precision.

XGBoost · Random Forest · SMOTE · model tuning ML / 03

COVID-19 Image Classification

Applies deep-learning image classification to labeled medical imagery and demonstrates the computer-vision workflow from data preparation through evaluation.

Computer vision · TensorFlow · Keras · CNN ML / 04

ReneWind Failure Prediction

Uses neural networks for predictive-maintenance classification, with iterative experiments that connect model behavior to an operational decision.

Neural networks · optimization · predictive maintenance ML / 05

Rental Price Model Deployment

Takes a serialized prediction model beyond the notebook with a REST backend, interactive frontend, container setup, and online and batch inference.

Flask · Streamlit · Docker · Hugging Face Spaces ML / 06

End-to-End ML Practice

Covers EDA, regression, classification, clustering, ensembles, feature engineering, NLP sentiment, article categorization, and model evaluation.

Pandas · scikit-learn · XGBoost · NLP · EDA

Explore all AI / ML projects

Production systems

Built for the messy middle.

The difficult part is rarely moving bytes. It is proving completeness, making retries safe, and giving operators a clear answer when a dependency fails.

SYS / 01Event driven

Batch completion detector

Moves orchestration from schedule-only assumptions to an event signal that validates the end record before downstream processing begins.

Azure FunctionsEvent HubsADFPython

SYS / 02Streaming

Warehouse event ingestion

Standardizes Kafka-compatible ingestion, durable checkpoints, schema evolution, partitioning, and recovery for high-frequency operational events.

DatabricksKafkaAuto LoaderDelta

SYS / 03Governance

Lakehouse migration

Maps legacy dependencies and moves data products toward Unity Catalog with repeatable development, validation, and production promotion.

Unity CatalogADLSMetadataCI/CD

Architecture explorer

Follow the decision path.

Switch views to see how the same engineering principles—clear contracts, idempotency, governed state, and observable failure—apply across different systems.

● validated pattern

DESIGN NOTE

Expertise

From source signal to trusted product.

I work across the full path, with a bias toward designs that stay understandable when the system is under pressure.

Technical range

StreamingEvent Hubs, Kafka, Auto Loader, Structured Streaming, checkpoints, replay, schema evolution

LakehouseDatabricks, PySpark, Delta Lake, Unity Catalog, medallion design, performance tuning

OrchestrationAzure Data Factory, Control-M, Azure Functions, metadata-driven pipelines, dependency handling

AI engineeringRAG, Vector Search, prompt engineering, MLflow tracing and evaluation, Model Serving, Genie, MCP, Lakebase

Machine learningscikit-learn, TensorFlow, Keras, XGBoost, feature engineering, ensembles, computer vision, NLP, model deployment

IntegrationService Bus, REST APIs, managed identity, Key Vault, private networking, operational alerts

Engineering principles

01Prove completeness before advancing the pipeline.
02Make retries safe, observable, and inexpensive.
03Separate environment configuration from business logic.
04Design failure paths as deliberately as happy paths.
05Turn operational knowledge into reusable platform patterns.

Connect

Let’s build data and AI systems people can trust.

I’m interested in Azure, Databricks, streaming, lakehouse, machine learning, and AI platform work where architecture, reliability, and business impact matter equally.

Open GitHub profile ↗