I work on AI/ML research and applied software projects, with a focus on building, testing, and evaluating intelligent systems.
My current work explores how machine learning models learn useful representations, align different types of data, and perform in retrieval-based tasks. I am especially interested in:
- 🧠 Multimodal learning
- 🔍 Retrieval systems
- 📊 Model evaluation
- 🧬 Representation learning
- 🏥 Medical AI
- 🛠️ Applied AI systems
I like building projects that do not stop at:
“The model trained successfully.”
I care about what the model actually learned, how it behaves, where it fails, and whether the evaluation proves meaningful progress.
|
|
|
|
|
A reliability-first framework integrating eBird, BBS, environmental covariates, and Earth-observation foundation-model representations with geographic, temporal, and uncertainty-aware evaluation. |
A controlled image-text contrastive retrieval study examining when cross-modal alignment forms, weakens, or fails under shifted and noisy conditions. |
|
A controlled benchmark studying how semantic false negatives affect contrastive retrieval and when false-negative-aware loss design improves representation quality. |
A domain-adversarial retrieval study testing whether reducing source-target representation shift preserves useful nearest-neighbor structure under visual distribution shifts. |
|
A multimodal retrieval benchmark exploring spectral geometry, sparse anchor supervision, InfoNCE, and MMD when exact cross-modal pairs are scarce. |
A diagnostic benchmark for studying embedding-space behavior through neighborhood preservation, graph connectivity, clustering, collapse indicators, and spectral structure. |
A natural-language movie discovery web application built with Next.js, TypeScript, Tailwind CSS, and movie APIs.
AI/ML: Python, PyTorch, NumPy, Pandas, Scikit-learn, Matplotlib
Software Development: TypeScript, Next.js, Tailwind CSS, Git, GitHub
Research Workflow: Experiment design, metric analysis, reproducible documentation, result interpretation
|
Controlled Experiments
Sample size, pairing, loss design, and split behavior |
Evaluation Metrics
Recall@K, lift-over-random, positive-pair similarity |
Embedding Analysis
Geometry, clustering, trustworthiness, spectral diagnostics |
Working Systems
Python ML pipelines, GitHub docs, Next.js web apps |
I am currently building a portfolio of AI/ML and applied software projects focused on model behavior, evaluation, retrieval, representation learning, and practical AI systems.
