Work

Studious is a learning-management system built for digital-first classrooms, and the longest-running thing on this programme. Alan founded it in September 2023 and still leads it: the frontend in Next.js, the backend on tRPC, and the infrastructure that keeps it responsive across regions — Pusher sockets for realtime, Supabase read replicas, load balancing on Render, and Vercel at the edge.

The Studious School Development Program grew out of it, bringing the platform to under-resourced classrooms, including two schools in rural Romania during a service trip in 2025.

Scribe is an AI study assistant that teaches method rather than handing over answers. Alan co-founded it and worked across the stack — he led the frontend redesign, built backend infrastructure, and supported the inference design behind its study workflows.

It took 1st Place Overall, the “Best Overall Hack”, at HackHarvard China 2025.

Sketch a chord progression or melody on notate’s piano roll and a decoder-only Transformer — six layers, 512-wide, eight heads, about 21 million parameters, trained on REMI-tokenised MIDI from POP909 and larger piano corpora — continues the phrase. Everything plays through a polyphonic subtractive synth Alan wrote from scratch as a Web Audio AudioWorklet; the same DSP renders the WAV export.

It sits where the two halves of this programme meet: a violinist’s ear for phrase, and an engineer’s interest in what a model can learn about it.

AI-native workspace that takes a physical-product idea from brief to CAD, PCB, verification, storefront, and checkout.

• Project-aware AI copilot with streaming tool execution and background jobs.
• Mechanical CAD (KCL / Zoo), schematic and PCB workspaces, Monaco editor with Yjs multiplayer.
• Verification gates, then v0-generated storefront and Shopify-backed checkout.

Campus planner plus student network — prerequisite graphs, dependency-aware plans, and people discovery.

• Interactive course planner with prerequisite dependency graphs.
• Social layer for finding classmates, communities, and relevant peers.

An AI messaging app that reads tone, emotion and intent — from voice, text and facial expression — and carries it with every message. Built at HackHarvard 2025.

• Voice messages transcribed with the emotion detected from tone, pitch and rhythm.
• Text messages tagged with emotion read from facial expression on the webcam.
• Any annotated message can be played back as expressive speech in the sender’s cloned voice.

Study groups that connect students across borders and languages, with an AI assistant in every group — built for Empower 3.0 Hacks.