Built a data-quality and ML pipeline for diabetes health tracking, improving evaluation readiness and supporting blood glucose spike detection with ROC-AUC 0.72–0.78.

• ›Designed a data-quality workflow to clean and validate health inputs (format, timestamps, completeness), producing analysis-ready time-series datasets that improved ML evaluation and reporting reliability.
• ›Developed Python pipelines to normalize time-series records and add anomaly checks plus triage summaries, improving evaluation readiness and supporting blood glucose spike detection with ROC-AUC 0.72–0.78.
• ›Implemented automated consistency checks to catch edge cases early and generate interpretable diagnostic summaries, improving output reliability and speeding debugging across model and data-pipeline iterations.

Service-oriented REST API backend with strict data validation, optimized SQL schemas, and automated quality checks to support reliable ML/analytics pipelines.

• ›Built service-oriented REST APIs with FastAPI and PostgreSQL, adding strict input validation, schema constraints, and debug tooling to improve data integrity and downstream ML/analytics readiness.
• ›Designed API contracts and optimized SQL schemas and queries to scale backend workflows, improve performance, and deliver consistent analytics-ready datasets across services.
• ›Implemented automated data-quality checks (nulls, duplicates, referential integrity) and KPI validation to reduce reporting errors and support reliable model inputs at scale.

End-to-end Parkinson's risk screening platform — voice audio in, 0–100 risk score out — powered by a Python ML inference pipeline trained on 200+ samples and served through a React frontend.

• ›Built a React/Vite frontend with authenticated requests, audio upload pipelines, and persistent result/history storage backed by a FastAPI + PostgreSQL service.
• ›Developed a Python ML inference pipeline using Librosa and Praat-Parselmouth for speech feature extraction, preprocessing, and binary Parkinson's risk prediction with a 0–100 score output.
• ›Trained and evaluated a TensorFlow classification model on 200+ voice samples, applying feature engineering and preprocessing to maximize predictive signal from raw audio data.

Full-stack web platform connecting local growers and buyers through a structured listing and messaging system, with validated data workflows and a responsive React interface.

• ›Built RESTful API endpoints with FastAPI and PostgreSQL, enforcing input validation, schema constraints, and referential integrity to keep listing and transaction data consistent.
• ›Designed a React frontend with dynamic filtering, listing creation flows, and real-time feedback, delivering a responsive experience across desktop and mobile viewports.
• ›Implemented data-quality checks and automated validation on user submissions to reduce malformed entries and improve downstream reporting accuracy.

Production-style multithreaded Python TCP client-server with structured logging, robust error handling, and fault-diagnosis tooling for concurrent workloads.

• ›Implemented a multithreaded Python TCP client-server with structured logging and robust error handling, debugging connection edge cases to improve stability and support production-like concurrent workloads.
• ›Added request validation, timeouts, and fault-diagnosis logs to handle unexpected inputs and failures, improving reliability and speeding troubleshooting with clearer runtime visibility during tests.