Portfolio
End-to-end machine learning projects — data ingestion, model training, evaluation and deployment — with real-world impact in health and human movement.
Building ML models that assess locomotive syndrome — age-related decline in mobility — from a single wearable IMU. Covers the full pipeline: raw accelerometer/gyroscope streams, signal cleaning, feature engineering from clinical movement tests, model training and validation against clinical gold standards. Published across two peer-reviewed papers.
Falls are a leading cause of injury in older adults. This project develops models that estimate fall risk from a single low-cost inertial sensor, replacing expensive lab-based gait analysis with screening that can run anywhere — a step toward preventive, at-home mobility care.
A deep learning model that simultaneously recognises quantitative (intensity) and qualitative (category) emotion by fusing visual data with physiological bio-signals. Explores multimodal fusion architectures and the trade-offs between unimodal and joint representations.
Active research investigating whether patterns in human movement, captured through multiple sensing modalities, can act as early digital biomarkers for dementia — combining motion capture, wearable sensing and machine learning.
Comprehensive performance evaluation of a novel postural balance assessment mat prototype, benchmarked against inertial sensors. Involved experimental design, synchronised multi-sensor data collection, and statistical analysis of agreement and reliability.
Reusable data pipelines and analytics workflows that cut manual processing time across research projects — automated ingestion of raw sensor data, cleaning and quality checks, reproducible training/evaluation runs, and visual reporting for non-technical stakeholders.
Deep learning models for image classification and object detection tasks — dataset curation and augmentation, transfer learning with modern CNN architectures, evaluation and error analysis, and deployment-ready packaging with OpenCV.
Open Source
Hands-on engineering projects with full source code — clone them, run them, break them.
Production-style Retrieval-Augmented Generation pipeline: LangChain orchestration, ChromaDB vector store, FastAPI REST API, four chunking strategies, similarity + MMR retrieval, conversational memory, and RAGAS quality evaluation. Fully Dockerised and unit-tested.
Three end-to-end analytical case studies — e-commerce sales, cohort retention, and health outcomes — each going from business question to SQL extraction (CTEs, window functions, cohort construction) to statistical testing in Python with DuckDB. Includes unit-tested stats utilities.
Desktop app for small businesses and freelancers: Tkinter GUI, branded PDF invoices with ReportLab, automatic numbering, GST and discount handling, CSV logging and formula-driven Excel export — with defensive error handling throughout.
That's exactly the kind of work I'm looking for.