An AI-powered tool designed to detect hallucinations in academic writing by cross-referencing generated claims against scholarly sources at scale.

Engineered a production full-stack system with Next.js 14 and FastAPI, integrating Gemini 2.0 Flash and a multi-layer validation pipeline across OpenAlex, Semantic Scholar, and Google Grounding. Launched to production and served 100+ users in its first week.

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

A research initiative investigating how eXplainable AI (XAI) influences user trust, empathy, and agency in emotional support LLMs.

Designed and built a custom A/B testing platform to compare baseline conversational models with systems augmented by sentiment analysis and transparency mechanisms. The project studies how explanation and user correction can reshape trust during high-stakes human-AI interaction.

Fields: HCI, Explainable AI, LLM Evaluation, User-Centred Design

A quantitative study modeling how psychological biases influence excessive trading behavior in financial markets.

Designed a cross-language data pipeline combining Python (data engineering, preprocessing) and R (PLS-SEM modeling) via rpy2, enabling statistical validation of behavioral hypotheses. The study demonstrates that Fear of Missing Out (FoMO) is a primary driver of irrational trading patterns.

Technologies: Python, Pandas, R, PLS-SEM, rpy2, Statistical Modeling

A custom academic markup language and IDE ecosystem designed to bridge Markdown-like ergonomics with structured publishing workflows.

Built the compiler frontend with ANTLR4, using AST traversal and the Visitor pattern to transform source documents into publishable representations. Developed the editing environment around extensible tooling, robust state management, and future-facing document compilation pipelines.

Technologies: Java, ANTLR4, JavaFX, MVVM, AST Design, Compiler Engineering

An experimental portfolio environment and personal web sandbox built for interaction design, frontend engineering, and developer-facing UX exploration.

Built with Next.js 15 and React 19, the site features a command-line-inspired interface, interactive motion systems, and canvas-driven visual effects. It functions as a living laboratory for interface experimentation, performance tuning, and product storytelling on the web.

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

Built in a high-intensity hackathon setting, Marketeer is an end-to-end AI marketing system generating full campaigns from a single prompt.

Designed an asynchronous multi-stage pipeline coordinating 5+ Gemini models to enable parallel asset generation. Implemented a browser-first workflow with IndexedDB persistence and FFmpeg WASM-based client-side media merging for fully in-browser production.

Technologies: Gemini API, TypeScript, IndexedDB, FFmpeg WASM, Async Pipelines

Built within a constrained hackathon timeframe, Solaura is a real-time assistive system designed to augment spatial awareness for visually impaired users.

Developed an iOS application using Swift, ARKit, and CoreML YOLOv8 for low-latency perception. Engineered a Python backend for state synchronization and audio synthesis, integrating Gemini and ElevenLabs for multimodal feedback.

Technologies: Swift, ARKit, CoreML, YOLOv8, Python, Gemini API, ElevenLabs

Developed in a hackathon setting, Hackonomics is an interactive economic simulation where players govern a village through cascading crises.

Built a custom macroeconomic simulation engine modeling GDP, CPI, unemployment, and dynamic event chains. Implemented bilingual narrative gameplay, persistent state management, achievement systems, and branching ideological endings without using an external game engine.

Technologies: Next.js 16, React 19, TypeScript, Tailwind CSS, localStorage, Game Systems Design

Built during a hackathon, Skillset LA generates personalized career roadmaps from user goals and background.

Engineered the backend system and integrated Gemini with external job-market data sources to transform fragmented inputs into structured, actionable development pathways.

Technologies: Python, FastAPI, Gemini API, Backend Systems

Built in a live hackathon environment, Realibuddy is a real-time assistant that verifies spoken claims during conversation.

Architected a low-latency backend pipeline integrating Deepgram ASR and retrieval systems via WebSockets to deliver grounded responses in real time.

Technologies: Python, WebSockets, Deepgram, Real-Time Systems