Selected Work

An AI-powered tool designed to detect "hallucinations" in academic writing by cross-referencing citations against OpenAlex (250M+ papers).

Engineered a Next.js 14 frontend and FastAPI backend. Features include a multi-tier verification system (OpenAlex, Semantic Scholar, Google Grounding) and a semantic auditor powered by Gemini 2.0 Flash. Served 100+ users in its first week.

Technologies: Next.js, FastAPI, Python, Tailwind, Google Gemini API

An immersive, terminal-themed portfolio designed exclusively for developers and tech enthusiasts.

Pushing the boundaries of Next.js 15 and React 19, this project features a fully interactive command-line interface (CLI), real-time "Matrix" rain simulations via Canvas, and hidden easter eggs. It serves as a living laboratory for my experiments in UI engineering and performance optimization.

Technologies: Next.js 15, React 19, TypeScript, Framer Motion, Canvas API

Winner of the Best AI/ML Award (Top 1 out of 29 teams). A real-time conversational assistant that verifies spoken claims instantly.

I architected the backend pipeline, integrating Deepgram for live transcription and Perplexity for grounded truth retrieval via WebSockets. The system achieved sub-second latency for real-time feedback.

Technologies: Python, WebSockets, Deepgram ASR, Perplexity API

Investigating how eXplainable AI (XAI) influences user trust in emotional support LLMs.

Designed an A/B testing platform to compare "black-box" models vs. transparent models with sentiment analysis. The study explores how user agency—the ability to correct AI predictions—can repair interactional breakdowns.

Fields: HCI, Explainable AI, LLMs, User-Centred Design

Quantitatively modelled the impact of psychological biases (FoMO, Herding) on excessive trading.

Architected a hybrid pipeline using Python (pandas) for data engineering and R (via rpy2) for PLS-SEM modelling. Validated that FoMO is a primary predictor of irrational market behaviour.

Technologies: Python, R, PLS-SEM, rpy2, Matplotlib

A desktop IDE for a custom markup language that blends Markdown simplicity with LaTeX precision.

Features an ANTLR4-based compiler front-end and a JavaFX UI built on MVVM architecture. Uses the Visitor pattern to convert ASTs to Pandoc-compatible JSON for professional publishing.

Technologies: Java, ANTLR4, JavaFX, MVVM, JUnit 5

A "Micro-SaaS" built in 4 hours to visualize academic goals.

Unlike traditional calculators, it employs Z-Score & Standard Deviation analysis to offer risk assessments on final exam targets. Features a modern Bento Grid UI, dark-pattern-free privacy (100% client-side), and custom i18n engine.

Technologies: Vanilla JS (ES6+), CSS3 Variables, Statistical Algorithms