AI / MACHINE LEARNING / BUILDING

Making data
work smarter.

I’m Zachary Steuer — an AI & machine learning practitioner turning complex problems into useful, thoughtful products.

BASED IN AUSTIN TXOPEN TO OPPORTUNITIES
Zachary Steuer smiling, in a gray suit and light blue tie
SCROLL TO EXPLORE

01 / SELECTED WORK

Ideas, in practice.

A selection of projects exploring how machine learning can solve real problems.

Medical Assistant Agent live demo: question box, covered guideline topics, and sources panel

GENERATIVE AI · RETRIEVAL-AUGMENTED GENERATION

Medical Assistant Agent

Built a clinical decision-support assistant that answers questions from 10 VA/DoD clinical practice guidelines and cites the guideline and page for every claim. It scored 4.38 / 5 on answer correctness against about 2.2 for a plain LLM, and declined every out-of-scope question. Includes hybrid retrieval, a React chat UI with a sources panel, an LLM-as-judge evaluation harness, Docker and CI. Educational project, not medical advice.

PythonRAGLLMsOpenAI APIGPT-4o-miniLangChainFastAPIChromaDBEmbeddingsHybrid Search (BM25)PyMuPDFReactTypeScriptViteVitestLLM EvaluationDockerNginxGitHub ActionsHugging Face Spaces
Design mockup of the coach live map: seven dinghies beating upwind with 5-minute trails, names and speeds, one boat flagged as last seen 22 seconds ago, race timer, and a boat details panel

MOBILE APP · REAL-TIME SPORTS ANALYTICS

Real-Time GPS Telemetry App

In-progress mobile MVP that streams live GPS position, speed, and heading to a coach’s tablet, with session replay and proprietary performance metrics in development. The image is a design mockup of the coach’s live map, built from the app’s spec. Field trials with a University of Texas athletics team begin in early 2027.

React NativeExpoTypeScriptFirebaseGPS & Location Tracking
ReneWind live demo: 86.5% failures caught, 96.4% alert precision, and turbine reading scorer

DEEP LEARNING · PREDICTIVE MAINTENANCE

ReneWind

Built a neural network that reads 40 anonymized sensor signals and flags wind-turbine generators likely to fail, so they can be repaired instead of replaced. Compared 8 network designs on heavily imbalanced data (5.5% failures), then tuned the alert threshold to minimize maintenance cost. On 5,000 held-out readings it catches 86.5% of failures with 96.4% alert precision (9 false alarms). Served through a FastAPI backend and a Streamlit demo, with tests, CI, and automated deploys.

PythonTensorFlowKerasNeural Networksscikit-learnpandasClass ImbalanceKNN ImputationCost-Sensitive ThresholdsFastAPIPydanticStreamlitpytestDockerGitHub ActionsHugging Face Spaces
Retail Sales Prediction live demo: model card comparing models, with the random forest at test R² 0.932 and 4.1% error

MACHINE LEARNING · RETAIL SALES PREDICTION & DEPLOYMENT

Retail Sales Prediction

Built a model that predicts how much revenue a product will generate in each store of a retail chain. Compared six models, from a store-average baseline to random forest and XGBoost, using cross-validation with leakage-free scikit-learn pipelines. The tuned random forest explains 93% of the variance on held-out data with about 4% average error, against 17% for the baseline. Served through a Flask API that rejects stores and values the model never saw, plus a Streamlit app, with tests, CI, and automated deploys.

Pythonpandasscikit-learnXGBoostRandom ForestGridSearchCVCross-ValidationML PipelinesRegressionPermutation ImportanceFlaskGunicornPydanticREST APIStreamlitpytestDockerDocker ComposeGitHub ActionsHugging Face Spaces
Loan Offer Response Prediction live demo: 97% hit rate, 89% recall, 10× lift, and customer scoring form

SUPERVISED LEARNING · EXPLAINABLE CLASSIFICATION

Loan Offer Response Prediction

Built an interpretable decision-tree model that estimates which existing bank customers will accept a personal loan offer, so marketing can prioritize outreach. Compared four trees with stratified cross-validation; when the selected model flags a customer it is right 97% of the time, about 10× better than random outreach, and it finds 89% of likely responders. The Streamlit app explains each estimate in plain language and includes an outreach planner that sizes the contact list. Marketing response only, not credit decisions.

PythonpandasNumPyscikit-learnDecision TreesCross-ValidationCost-Complexity PruningExplainable AILift & Gains AnalysisMatplotlibSeabornStreamlitpytestDockerGitHub ActionsHugging Face Spaces
FoodHub data case study: 1,898 orders across 178 restaurants and 14 cuisines

DATA ANALYTICS · EXPLORATORY ANALYSIS

FoodHub

Analyzed food delivery orders to understand restaurant and cuisine demand, customer ratings, order costs, and preparation and delivery times—turning the patterns into practical ideas for improving customer experience and operations.

PythonPandasNumPyMatplotlibSeabornJupyterExploratory Data AnalysisData VisualizationGitHub Pages

✳ Selected projects across machine learning, deployment, data analytics, and mobile software.

OPERATIONS EXPERIENCE

BUSINESS & PROGRAM LEADERSHIP · AUSTIN YACHT CLUB, AUG 2023–NOV 2026

Director, Program and Operations | Austin Yacht Club

Promoted in January 2026 after more than two years as Program Operations Manager. Directed program operations for a 450-family private membership club, owning $500K in annual program revenue (about 36% of the club’s $1.4M total revenue) and reporting P&L directly to the Board of Directors. Managed a 15-person staff end to end, including hiring, payroll setup, timekeeping, and performance reviews, and reduced payroll costs year over year in every department I oversaw. Grew fall program revenue 125% (from $20K to $45K), doubled enrollment from 60 to 120, and launched athletics programs with 8 Austin schools.

Program Operations Manager (Aug 2023–Dec 2025): Grew event participation 222% in the core youth segment (ages 8–13) and raised its retention from 30% to 92% over 3 years; overall retention reached about 90%, compared with about 52% across peer clubs. Expanded summer capacity from 25 to 30 participants per week while cutting on-duty staff from 7 to 4, adding $30K in revenue and saving $17,280 in payroll over an 8-week season.

Led Clubspot CRM adoption from launch and trained all staff, advised Clubspot on product feedback, rebuilt the club’s WordPress/PHP website, and built a churn prediction model to flag members at risk of leaving.

Visit website ↗

RELATED WORKFLOW EXAMPLES

  • Match invoice records with payment records and flag discrepancies.
  • Surface overdue or incomplete invoices for follow-up.
  • Create reminders and follow-up tasks for member or account workflows.
  • Automate routine CRM record updates and status tracking.
Staff LeadershipProgram OperationsP&L ReportingPayroll & HiringWordPressClubspot CRMInvoice ReconciliationChurn PredictionWorkflow Automation
02THE PERSON
BEHIND THE MODELS

A LITTLE ABOUT ME

Curious by nature.
Practical by design.

I like the space between a promising idea and something people can actually use.

I work across the machine learning process: exploring data, building and evaluating models, and finding the right way to put useful results in people’s hands. I care about clear thinking, honest evaluation, and making the complicated feel approachable.

A bit more about me ↗
CURRENTLY
Graduate Certificate in AI & Machine Learning: Business Applications, McCombs School of Business, The University of Texas at Austin, 2026.
5 APPLIED PROJECTS · 8 MONTHS✳

03 / THE TOOLBOX

Tools for the curious.

My current focus is AI, machine learning, and data analytics, backed by hands-on work in APIs, databases, deployment, and full stack and mobile development.

✳

Current work centers on data, predictive modeling, and neural networks, from exploratory analysis to a tuned, evaluated model.

MACHINE LEARNING & DATA SCIENCE · CURRENT

PythonPandasNumPyscikit-learnTensorFlowKerasXGBoostJupyterGoogle ColabMatplotlibSeabornExploratory Data AnalysisUnivariate ProfilingApplied StatisticsClassificationRegressionFeedforward Neural NetworksDecision TreesRandom ForestGradient BoostingLogistic RegressionBagging & StackingImbalanced Data & Class WeightsKNN ImputationFeature ScalingML PipelinesCross-ValidationHyperparameter TuningGridSearchCVCost Complexity PruningCost-Sensitive Threshold TuningPermutation ImportanceExplainable AI (Decision Paths)Lift & Gains AnalysisModel Evaluation
◎

Applications built on large language models, measured against real answers rather than impressions.

GENERATIVE AI & LLMS · CURRENT

Retrieval-Augmented Generation (RAG)OpenAI APIAnthropic Claude APILangChainChromaDBVector EmbeddingsSentence TransformersHybrid Search (BM25 + Dense)MMR Re-rankingPrompt EngineeringCitation GroundingLLM-as-Judge EvaluationMulti-Agent PipelinestiktokenPDF Parsing (PyMuPDF)
⇄

Turning models and automations into tested services that other people can call.

DEPLOYMENT, APIS & ENGINEERING · CURRENT

Model DeploymentFastAPIFlaskGunicornStreamlitPydanticREST APIsThird-Party API IntegrationOAuth 2.0QuickBooks Online APIWeb Scraping (BeautifulSoup)Job Scheduling (APScheduler)DockerDocker ComposeHugging Face SpacesGitHub Actions CI/CDpytestRuffmypyuvStructured LoggingLinuxGitGitHubGitHub CodespacesGitHub Integration
▤

Relational, document, embedded, and vector databases, used across current projects and earlier full stack work.

DATABASES & DATA STORAGE

SQLMySQLSchema Design & SeedingSequelize ORMMongoDBMongooseSQLiteFirebase FirestoreFirebase Realtime DatabaseChromaDB (Vector Database)IndexedDBCRUD Operations
◇

Typed web and mobile front ends for the people who actually use the tools.

FRONTEND & MOBILE · CURRENT

TypeScriptReactViteReact NativeExpoFirebase AuthFirebase Cloud FunctionsPush NotificationsGPS & Location TrackingMapLibreJestVitestReact Testing LibraryESLintPrettierAccessibility (a11y)Monorepos (npm Workspaces)
↗

Earlier training in web application development through a JavaScript full stack bootcamp in 2022.

FULL STACK DEVELOPMENT · BOOTCAMP, 2022

HTMLCSSJavaScriptReactNode.jsExpress.jsnpmDOM ManipulationjQueryREST APIsGraphQLApolloJSONMVC ArchitectureHandlebarsResponsive DesignBootstrapMaterial UIFrontend-Backend IntegrationSQLMySQLSequelizeMongoDBMongooseCRUD OperationsAuthenticationJWTbcryptSessions & CookiesProgressive Web AppsService WorkersWebpackJestGitGitHubDeployment

HAVE A GOOD QUESTION?

Let’s make
something useful.

Say hello↗