At SkillArion Technologies Pvt. Ltd., I trained a five-class dog emotion classifier and re-evaluated its results after finding that breed-folder recording cues inflated the reported score. I corrected the accuracy from 93% to 88% before it reached stakeholders.
I deployed the models behind a Flask API on AWS EC2 with a Streamlit demo, working with an ML engineer, the hardware team and a data lead. I also reported per-class results and model limitations to non-technical stakeholders.
On my SQL Order Audit project, I reconciled payments against 99,441 orders and found 772 paid orders with no items recorded, worth R$162.6K. I investigated apparent duplicate payments and packaged the checks as reusable SQL views.
I built DataRefinery to clean CSV and Excel datasets in a single run and report missing values. In my Cafe Financial Dashboard project, I audited transaction data and built Power BI dashboards for campaign ROI and cash flow.
