Selected into the prestigious Summer Science Program (SSP), I worked on an SSP-led bacterial-genomincs research project at the University of Guelph that combined wet-lab genomics (DNA extraction, PCR, library prep) with computational analysis (read trimming → assembly, mutation calling, comparative genomics). The study included a cross-dataset meta-analysis with collaborating groups, and my results contribute to a forthcoming SSP group manuscript. Practical goal: identify resistance-linked mutations to monitor for early surveillance of drug-resistant strains.

Open-source, resource-efficient RNA tertiary-structure pipeline that combines a RoPE-enhanced Transformer encoder with an EGNN refinement stage. Built to run accurate RNA 3D predictions on a single GPU—via careful memory/compute optimizations (attention approximations, slimmer MLPs, residual MLP refinements). Benchmarked with reproducible scripts and small pre-trained weights; shows lower RMSD vs. lightweight baselines.

Open, reproducible pipeline that reframes LLM “hallucinations” into useful biomedical ideation. The system automates prompting → candidate generation → scoring → light validation, using targeted prompt strategies and low-cost checks to encourage novelty while filtering obvious errors. Across multiple models and prompt variants it generated and curated ~400–500 candidate concepts, with reproducible metrics + human review showing more actionable leads than naïve prompting— all compute- and budget-friendly and designed with safety guardrails.

A compact, practical intervention that pairs low-intensity aerobic training (MAF-180) with a structured Yoga Namaskar / posture program to counter spine pain and mood issues linked to prolonged social-media use. I designed and ran the independent study (participants, protocol, IRB-style documentation), collected pre/post quantitative measures (posture, pain scores, mobility) and subjective outcomes (energy, mood), and used basic stats to show consistent improvements across participants. Result: a low-cost toolkit appropriate for schools, community programs, and telehealth.

A three-year, real-world training log that tracks how consistent low-intensity (MAF-180) workouts change aerobic efficiency in a single-subject case study (South-Asian Male). I collected a longitudinal dataset (heart-rate, pace, adherence), ran heart-rate & pace analytics, and used statistical comparisons to show faster pace and better recovery at the same heart rate—while controlling for training volume and consistency. All methods were documented for reproducibility (analysis scripts prepared). The manuscript is accepted pending revisions by the Journal of Emerging Investigators (JEI).

A collaborative clinical study with ophthalmology specialists and a renowned Ortho-K expert, leveraging a multi-year real patient dataset to develop a practical early-screening workflow that predicts whether a candidate will respond successfully to overnight Orthokeratology (Ortho-K) lenses. Instead of relying on long, costly trial-and-error, the project focuses on baseline and short-term signals — corneal topography/shape, axial-length trends, lens centration/fit, corneal biomechanics, tear-film quality, and simple compliance/behavioral indicators. A secure dataset and analysis pipeline were built to correlate these features with early lens response, and we are piloting scoring rules and exploratory models to flag likely-success candidates within weeks rather than months.