SalmanMKC

Silent clip. Meta Quest headset footage first, then the desktop simulator.

Roomcraft — author a whole virtual place together by describing it

An opt-in XR Blocks addon that turns a prompt into editable 3D content, composing catalog assets with procedural primitive-part designs. Scenes are refined in place through text, Gemini voice or direct manipulation — alone, or with several people sharing one room code across headsets, phones and laptops. Optional virtual worlds, characters and creatures set moving, bounded animation, undo/redo and JSON export.

Silent clip.

AgentHands — speech-synchronised gestures and room-aware pointing

Translucent posable hands and a glowing orb that gesture while an AI agent speaks. Gesture markup is synchronised to real text-to-speech word boundaries, and the agent points at objects in the room, grounded through depth. Runs scripted with no API key, or live with Gemini.

Silent clip, recorded in the desktop simulator.

Persistent Spatial Anchors — content that stays put

Capability-aware WebXR anchors that keep content tied to a physical location across sessions. On Meta Quest the platform handles real anchor persistence; elsewhere the addon falls back to session-only anchors, and the desktop simulator keeps the demo working without a headset.

Silent clip: two devices in one session.

Multiplayer — shared mixed reality across devices

Multiplayer networking for XR Blocks: proximity voice, shared chat, shared movable objects, RPC actions, presence and three transport protocols, across networks, custom servers or locally between Meta Quest, Android XR, mobile and PC.

MCP Server for Coding Agents — Google XR Blocks

A stdio MCP server that exposes list_skills, get_skill and search_api so coding agents can discover XR Blocks workflows and check real exported APIs instead of inventing them. A follow-up extended the API surface to addons, so answers stay accurate as the SDK grows.

Audio-Driven Avatar Lipsync

End-to-end audio-driven facial animation that runs locally in the browser. Mic → MFCCs → small MLP → 52 ARKit blendshapes → stylised avatar. The browser trains the model itself with TF.js on data captured live from your face via MediaPipe FaceLandmarker. The same architecture is mirrored in a JAX/Flax trainer that exports to LiteRT for Android XR or Quest.

On-Device Language Detection + Gemini Voice — Google XR Blocks

Spartial-transcript pipeline that detects language per utterance. Uses Gemini Live for the audio stream and MediaPipe's language detector on device to tag each utterance. Lays out as a chat-style list with the latest at the bottom — slots scale together so it doesn't jump when a new line comes in.

Metaverse AR Therapy (CHI 2025, Japan)

Research project exploring AR-based therapeutic experiences in the metaverse — published at CHI 2025.

Microsoft Patent — Emphasis Image Generation

Inventor on a Microsoft patent application: "Generation of Emphasis Image with Emphasis Boundary".

Portals Demo — Google XR Blocks

A showcase XR experience built on XR Blocks demonstrating immersive scene composition and interaction primitives.

Gamepad Controller Support — Google XR Blocks

Added gamepad input support to XR Blocks so developers can build XR apps that work with standard controllers.

Youth WellTech (arXiv, with Microsoft Research)

A global remote co-design sprint for youth mental health technology, with Microsoft Research. I was a mentor and the second-highest contributor on the paper.

UCL Agentverse Hackathon #1 — Multi-Agent System

Winning multi-agent system built with CAIPE / LangGraph at the UCL Agentverse Hackathon.

Mixed Reality Smart Office — HoloLens

HoloLens experience visualising IoT data in 3D with dynamic image tracking on physical devices around the office.