I build practical tools where machine learning meets real people. Each project below is pinned up like evidence: the problem, what was built, the judgment call, and the honest status.
The system watches through a webcam, finds the face, and classifies its expression with an emotion model from the FER-2013 family. Raw predictions flicker, so the mood is decided by a majority vote over a sliding window, with a minimum hold time before the music may change — then tracks crossfade, never cut.
It exists in two forms: a hands-free Python desktop app (caregivers drop the patient's favourite songs into mood folders — familiar music works best) and a browser demo that runs entirely on your device and synthesizes its music live, so anyone can test it without installing anything.
The project began during a fellowship at Samsung Electronics America, exploring facial-emotion recognition for Alzheimer's care.
Generalized from a real workflow: utility-billing remittances pulled from SQL Server, reconciled by hand in Excel while the book changed underneath — payments posting, services cancelling, adjustments landing. This rebuild uses 100% synthetic, self-generated data and no employer code or rules.
Per account, every cycle: three detail-to-summary sum checks plus an R-C identity check against what the export claims. When duplicate dollar amounts inflate a column, the engine zeroes the duplicate occurrences — never a real amount, never a delete — until detail ties to summary; every repair carries a SOX-style adjustment note for manager approval, and anything unfixable is routed to manual review.
The pipeline runs itself: every 30 minutes a GitHub Action appends a new synthetic remit cycle, reruns all 18 dbt models and 43 tests, commits the refreshed warehouse, and the public app redeploys. The app includes a workbench that replays any account's reconciliation step by step, and a pipeline page that renders the lineage graph and test results from dbt's own artifacts.
Business Analyst — Retail Operations, Constellation Energy (contract), Houston · 2025–2026. Billing application workflows for commercial energy accounts: requirements, UAT, root-cause analysis, and Python automation for remittance reconciliation. UAT work contributed to a ~30% reduction in billing errors.
IT Business Systems Analyst, Resource Environmental Solutions · 2024–2025. Salesforce and NetSuite configuration, Power BI validation, requirements for a OneStream implementation (working with CFO, VPs, Directors), and Python pipelines feeding Microsoft Fabric.
Technical Solutions Analyst, GoDaddy · Asurion · Lexia Learning · 2022–2024. Frontline software and systems support; knowledge-base documentation that cut repeat escalations by 15%.
Earlier: Operations Supervisor at CVS Health (promoted within 6 months; proposed the store's dedicated COVID-vaccination kiosk), and Healthcare Support Administrator at Atwell Home Healthcare — EHR administration (EpicCare Home Health, CradleMRx, Kantime, OASIS) and a patient census grown from 10 to 50 in under a year.
Education: M.S. Analytics, Georgia Institute of Technology (in progress, 2027) · MBA in Management Information Systems, Lamar University (2023) · B.S. Health Sciences, Herzing University (2021). Machine-learning fellowship at Samsung Electronics America — where Case 001 began.
Business analyst who builds practical tools where data meets real people — analysis, automation, and applied machine learning. A decade of experience across healthcare operations, technical support, and business systems (Constellation Energy, Resource Environmental Solutions, GoDaddy/Asurion/Lexia Learning, CVS Health, Atwell Home Healthcare).
Education: M.S. Analytics, Georgia Institute of Technology (in progress, 2027) · MBA in Management Information Systems, Lamar University (2023) · B.S. Health Sciences, Herzing University (2021). Machine-learning fellowship at Samsung Electronics America.
Tools: SQL, Python, Power BI, Tableau, Microsoft Fabric, Salesforce, NetSuite, Excel (Power Query, VBA), EHR systems (EpicCare Home Health, CradleMRx, Kantime, OASIS).
joan_elendu@yahoo.com · linkedin.com/in/joanelendu · github.com/jelendu