Experience
- Present
AI Researcher
May 2025 – Present
Applied NLP and ML to socio-economic Twitter data to build financial stress analytics.Highlights
- Built an end-to-end NLP pipeline over a 10-year Twitter corpus (ingestion, cleaning, labeling, evaluation) to produce a Financial Stress Index (FSI) for Canada.
- Engineered reproducible data workflows (documented filters, regex rules, deterministic seeds, versioned datasets) to ensure high-quality tweet selection and preprocessing.
- Designed hybrid sentiment and feature extraction using lexicons and LLMs (BERT & VADER) to score financial polarity and intensity with quality controls.
-
Graduate Research & Teaching Assistant
Sep 2023 – Dec 2024
Research and teaching support in federated learning, optimization, and course assistance.Highlights
- Reviewed challenges of federated learning in IoT networks and benchmarked FedAvg/FedProx/SCAFFOLD on CICIoT2023 under statistical heterogeneity.
- Proposed a deep learning training optimization (REDUS) reducing training time by ~72% with ~1.6% accuracy trade-off.
- Implemented automated testing frameworks for grading programming/networking courses using C++/Python/Docker.
-
Intern Researcher
Centre for Informatics Sciences
Feb 2022 – May 2023
Medical imaging research in Alzheimer’s detection and breast cancer segmentation.Highlights
- Conducted two applied AI studies: Alzheimer’s detection (multimodal features) and breast-cancer ultrasound segmentation (U-Net in a federated setting).
- Addressed class imbalance using SMOTE and improved a published Alzheimer’s baseline by ~10%. Achieved ~96% segmentation accuracy for BC in FL and published at MIUA.
- Documented preprocessing, imbalance handling, and evaluation (Dice/IoU/ROC-AUC) with reproducible training scripts and clear experiment logs.
-
Intern AI & Embedded Software Developer
Delta Care
Jun 2021 – Sep 2021
Developed embedded and AI components for reproductive health analysis systems.Highlights
- Implemented a C-based PID temperature controller for precise sample regulation in a lab workflow.
- Optimized Python-C/C++ interprocess communication, reducing response latency by ~70%.
- Built CV pipelines: YOLOv5 + DeepSORT for motility tracking (~91% accuracy) and Mask R-CNN for morphology segmentation (~87% accuracy).