Portfolio

Won 3rd place in the Google DeepMind track at the {Tech: Europe} London Hackathon. SAI is a background AI agent that shops, tries on clothes, and pays on your behalf - Siri, but it actually gets things done. Powered by the Gemini 3.1 Flash Live API, it browses a luxury storefront through real-time voice conversation, generates virtual try-ons from your camera and product photos via fal.ai, and completes checkout with an AP2-enabled merchant. Under the hood, payments execute through a cryptographically signed mandate chain (Intent → Cart → Payment) built on Google's AP2 (Agent Payments Protocol) - an open trust layer for agent commerce where every step is verifiable and every transaction auditable. A real glimpse of where agentic commerce is heading.

I built Allos, a full-stack analytics platform for merger arbitrage investors that automates deal discovery by scanning SEC EDGAR 8-K filings and extracting terms with an LLM. It computes live financial metrics - gross spread, annualized IRR, implied probability, and beta-adjusted downside - across cash, stock, and mixed deal structures. The platform features a sortable universe dashboard, per-deal detail pages with editable terms and regulatory risk tracking, and an async job system for background metric recalculation. Built with Python/FastAPI, PostgreSQL, Next.js 16, TypeScript, Redux Toolkit, and Shadcn UI.

I participated in the FLUX.2 Hack in London, organised by Black Forest Labs, and built InteriorMaxi over 48 hours at Dawn Capital. The application allows users to upload a floor plan, specify their style preferences, and receive photorealistic renders for every room — then shop the exact furniture. The trickiest technical challenge was getting FLUX.2 to maintain visual consistency across rooms within an apartment. Built with access to BFL's latest models and mentorship from Anthropic and NVIDIA, this project demonstrated my ability to rapidly prototype AI-powered products and collaborate effectively under hackathon constraints.

I participated in the Granola v0 Hackathon to build VibeCast, a Next.js application that transforms meeting transcripts into tone-aware, storyboarded video recaps. The project featured transcript input with configurable tone presets, AI-powered storyboard generation with shot breakdown visualization, and a marketing site, utilizing Next.js 14, TypeScript, Tailwind CSS, and Shadcn UI. As a result, I strengthened my ability to deliver demo-ready experiences under time constraints.

Having mastered full-stack development through various projects, I wanted to explore the intersection of 2D interaction design and 3D/AR experiences. At London VibeHack, I built RoboVibe, a sketch-to-3D prototyping platform that allows users to draw robot silhouettes in a browser and instantly convert them into AR-ready 3D models, achieving 3rd place. The complete pipeline - from canvas drawing to AWS S3 storage, QR code generation, and AR visualization - was engineered in less than 10 hours using Next.js 16, React 19, FastAPI, Three.js, and fal.ai's TripoSR model. This experience demonstrated my capability to rapidly orchestrate complex full-stack workflows and integrate cutting-edge AI services into production-ready applications.

• Built and managed a React.js and TypeScript-based Docker containerised e-commerce ads builder, enhancing campaign efficiency and achieving business needs.
• Designed and implemented API client library with security controls embedded to simplify API calls for developers and improve DX.
• Researched, coded, managed and tested a centralized documentation and integrations portal with Next.js, MDX, and TypeScript, improving accessibility to up-to-date information.
• Designed and maintained front-end infrastructure on AWS using Lambda@Edge, S3 wildcard subdomains, and CloudFront using Terraform, leading to improved DX and build process automation.
• Developed automated unit and integration tests using Vitest, React Testing library and Cypress, ensuring robustness and optimal system performance.
• Planned and led development on agile stories, assisting colleagues and communicating with vested stakeholders to provide feedback and iterate what needs to be built.