Publications, writing, and code.

AI-Driven Early-Warning System to Predict Multi-Organ Deterioration in Critical Care Patients

Temporal deep learning with cross-organ attention to detect ICU deterioration a median of 6.2 hours before conventional clinical detection, across five organ systems. Presented at IEEE ICHI 2026, Session 16. Co-authored with Sarmista Thalapaneni; session chaired by Dr. Chen-Ren Shu, PhD, FACMI, FAMIA.

AI-driven drug repurposing: a graph neural network and self-supervised learning approach

Computational drug discovery using GNNs and self-supervised pretraining over biomedical knowledge graphs.

Multi-modality medical image fusion using ML and deep learning

Methods for integrating CT, MRI, and ultrasound modalities for clinical decision support. Book chapter.

Reinforcement learning for decision making

Applications of RL across modern Industry 5.0 systems. Book chapter for CRC Press (Taylor & Francis).

Do trust-based social recommendation algorithms work as intended?

Empirical study examining whether trust signals deliver on their promise in recommender systems.

Influence of social circles on user recommendations

Master's thesis on social-graph-aware recommendation models. San Jose State University.

Machine translation of English videos to Indian regional languages

End-to-end pipeline for translating spoken English video into regional Indian languages.

cloudfit

Cloud-agnostic machine type recommender for batch and bioinformatics workloads. Given a workload spec (CPU, RAM, region, optimize for cost / performance / availability), returns ranked instance recommendations with transparent per-factor scoring. Multi-package OSS ecosystem: scoring engine (cloudfit-core), GCP provider (cloudfit-provider-gcp), and a stateless FastAPI service (cloudfit-api) with a multi-region bundled snapshot. Built to fill the pre-launch and batch-workload sizing gap that incumbent free tools (Compute Optimizer, GCP Recommender, Azure Advisor) don't cover.

$ pip install cloudfit-core cloudfit-provider-gcp

clinops

Clinical ML pipeline toolkit: MIMIC-IV / FHIR loaders, temporal feature windows, and patient-aware train/test splits that don't leak across cohorts. Distilled from production work in clinical and genomic data engineering.

$ pip install clinops

samplesheet-parser

Format-agnostic parser for Illumina SampleSheet.csv files. Auto-detects IEM v1 vs. BCLConvert v2, validates index integrity with Hamming distance checks, and converts, diffs, or merges sheets across mixed sequencing fleets.

$ pip install samplesheet-parser

Why I built cloudfit

Launch post. The gap between free cloud sizing tools (Compute Optimizer, GCP Recommender, Azure Advisor) and what I actually needed for new batch and bioinformatics workloads. Why an open-source recommender that doesn't need historical metrics from a running workload.

Understanding recommender systems: the engine behind personalized experiences

A primer on collaborative filtering, content-based, and hybrid approaches to recommendation. Why personalization engines work the way they do, and where they break.

An introduction to explainable AI and explainable boosting machines

A primer on XAI fundamentals and how EBMs combine accuracy with interpretability.

Build kNN from scratch in Python

A from-the-ground-up implementation of k-nearest neighbors with the math made explicit.

Understanding the NLP pipeline

A walk-through of the typical stages in a modern natural-language-processing system.

Intro to ETL pipelines

A walk-through of extract-transform-load patterns and why they matter for data platforms.