Toronto, CanadaOpen to work

Hi, I'm Aman Shah 👋

CS Student. SWE & MLE.

Recently, I've been building fullstack web apps, accessibility prototypes, and games. I love documenting my process and making friends along the way (seriously lol).

SWE / MLE

Hobbyist game developer

YouTuber building in public

Computer Science @ UofT

See projects Open resume Get in touch

Selected work

Project stories from start to finish (or kinda finished i guess 😅).

Side projects, hackathon wins, prototypes, and way too many late nights.

agentson-device sensingdesktop apps

Recompile

Developer-wellness companion that waits for the commit

Reads your vitals through the webcam and your work through git and calendar, fuses both into one cognitive-load score, then waits for a commit to interrupt.

When: July 2026
Role: Full-stack engineer, agent core, sensing pipeline, load model, desktop app, design system
Context: Built for Hack the 6ix 2026. One always-on Electron tray process is the brain; every window is a thin client with no independent state, so closing them all changes nothing.

Why it mattered

Every break timer interrupts on a clock. When a stressed person should step away, and what would help, is a judgment problem rather than a rules problem, so an agent core makes the call.

Outcome

Winner

Won Most Innovative Use of Backboard.io at Hack the 6ix with a live loop: webcam vitals, a cognitive-load score, boundary-held nudges, push-to-talk voice, and a painted tree grown from your days.

Hack the 6ix, Most Innovative Use of Backboard.io

Gated every sensing path behind one ~37-line consent check, with rPPG kept on-device.
Scored load from four signals, dropping and renormalizing any term that goes stale.
Queued nudges instead of firing them, holding until a commit or merged PR lands.

TypeScriptElectronReactMongoDB AtlasBackboard

3D graphicsdata reconstructionreal-time systems

PitchSide

A time machine for football, in the browser

Rebuilds ~4,000 real matches from their recorded event streams into 3D worlds you can pause, rewind, fly through, and watch through a player's eyes.

When: July 2026
Role: Solo engineer, data reconstruction, 3D runtime, cameras, audio, product design
Context: Built for United Hacks V7. Runs entirely client-side: no backend beyond one optional LLM route, no database, no accounts.

Why it mattered

Event data says who passed from where to where, but never where all 22 players stood at each instant. The work is making that sparse record continuous without ever contradicting it.

Outcome

Winner

Won 1st place at United Hacks V7 with a browser match player spanning Pelé's 1958 final to Messi's 2022, carrying five camera modes, an auto-cutting director, and per-decade film stocks.

United Hacks V7, 1st place

Stitched the ball path from real waypoints, hashing each player's id into a persona.
Normalized each match into uniform-grid typed arrays so a seek is one O(1) sample.
Drove mixers, gait, and audio by match time, so pause freezes mid-stride for free.

GitHub

TypeScriptNext.jsthree.jsReact Three FiberAnthropic SDK

AI producttravelmaps

NexTrip

AI travel planning web app

Builds tailored vacations through a conversational planning flow, then turns preferences into day-by-day plans with maps and saved itineraries.

When: March 2026
Role: Frontend engineering, planning flow, map experience
Context: Built for GenAI Genesis 2026 at UofT.

Why it mattered

Travel planning is usually split across chat, maps, search, and documents. NexTrip keeps the preference gathering and itinerary building in one workflow.

Outcome

Delivered a working itinerary builder with chat-mode planning, interactive maps, and saved trip state.

Designed a chat flow that collects interests, restrictions, and planning constraints.
Generated structured daily schedules rather than loose recommendations.
Connected itinerary output to map interactions for spatial planning.

GitHub Devpost

TypeScriptNext.jsGemini APILeaflet API

AI producteducationfull-stack

Synapse

Google Classroom-style AI study platform

Connects teacher dashboards with adaptive student study modes, turning uploaded class materials into chat, notes, mind maps, games, flashcards, podcasts, and quizzes.

When: 2026
Role: Full-stack engineer, AI workflow design, education product design
Context: Full-stack education platform built around teachers, students, class material, and AI-assisted studying.

Why it mattered

Most classroom tools separate content management from actual studying. Synapse closes that loop so teachers can see how students are learning while students get help that responds to their weak spots.

Outcome

Winner

Won 3rd place at the Agent Shyft Hackathon during Toronto Tech Week, with a two-sided learning experience where student activity surfaces progress and weak spots back to teachers.

Agent Shyft Hackathon, 3rd place

Created a teacher workflow for uploading learning materials and monitoring student progress.
Designed student study modes across smart notes, mind maps, story games, flashcards, podcasts, chat, and quizzes.
Used quiz performance to guide AI responses toward each student's weaker topics.

TypeScriptNext.jsAI chatAdaptive quizzesEducation UX

agentscommercehackathon winner

Haggle

AI agents that negotiate product deals by phone

Finds sellers for a product, scrapes contact information, sends calling agents to negotiate, and returns the strongest deals it can find.

When: 2026
Role: Full-stack engineer, agent workflow design, web automation
Context: Winner at the Cursor Hackathon during Toronto Tech Week.

Why it mattered

Getting a good deal is time-consuming because it means searching, calling, comparing, and negotiating across sellers. Haggle turns that into an agent-led workflow.

Outcome

Winner

Won 2nd place at the Cursor Hackathon during Toronto Tech Week with a working flow that searched for sellers, contacted them, and summarized negotiated offers for the user.

Cursor Hackathon, 2nd place

Built a request flow where users ask for the best available deal on a product.
Scraped the web for seller phone numbers and contact paths.
Returned negotiated options so users could compare the best deals without making every call themselves.

GitHub LinkedIn

TypeScriptNext.jsWeb scrapingVoice agentsAutomation

frontendhealth researchpublic awareness

Fontan Research

Public research and awareness website

A frontend website created to spread awareness about Fontan circulation and share Fontan research led by Dr. Ashish Shah at St. Boniface Hospital.

When: 2026
Role: Frontend developer, information design, public health content structure
Context: Built as a public-facing resource for Fontan education, research visibility, and patient-family awareness.

Why it mattered

Fontan physiology is complex and often unfamiliar outside specialist care. The site gives the topic a clearer public entry point and makes ongoing research easier to find.

Outcome

Commercial

Published a live website that organizes Fontan background, research context, and hospital-linked work into a more approachable web experience.

Public website for Fontan research awareness

Structured medical and research information for non-specialist readers.
Built a calm, readable frontend around patient-family awareness and research credibility.
Created a deployed public resource connected to Fontan research at St. Boniface Hospital.

Live

Next.jsTypeScriptResponsive UIContent designVercel

AI producteducationaccessibility

Flux

AI study system for neurodivergent students

Turns lectures, PDFs, videos, and notes into adaptive study sessions with podcasts, games, flashcards, quizzes, and smart notes.

When: March 2026
Role: Full-stack engineer, product design, AI workflow design
Context: Built for the Google Developer Group's Build with AI case competition.

Why it mattered

Many study tools assume one learning mode. Flux was designed around switching formats so students can learn through listening, recall, practice, and play.

Outcome

Winner

Won 1st place at the Google Developer Group Build with AI case competition, and became the clearest example of how I approach product thinking, accessibility, and applied AI in one build.

GDG Build with AI, 1st place

Designed an ingestion flow for audio, video, PDF, and written study material.
Mapped source material into various study modes instead of a single summary output.
Built the product around neurodivergent learning needs, pacing, and variety.

GitHub LinkedIn

TypeScriptNext.jsCerebras APIPrismaAI pipelines

Experience

Learning, Working, Shipping.

Interested in big tech, startups, and making my own creative projects (check out my indie game 🥹).

June 2026 to present
AI Software Engineer
KnowIdea
Internship
Developing predictive intelligence platform for businesses. Optimizing generation pipeline (cost/time), maintaining 500K+ ARR, and automating client work.
Location
Toronto
500K ARR
May 2026 to present
ML Engineer
HeartWise
Internship
Building an image augmentation pipeline to expand training data for a medical imaging model, improving performance on smartphone ECG images. Complementing current model performance on digital ECG files.
Location
Toronto
AI/ML1.9M+ data pointshealth tech
Aug 2021 to present
Game Developer
TheUnityGuy
Developing a commercial first-person shooter while building custom Unity tools, environments, combat systems, and public devlogs around the process.
Location
Remote
UnityC#custom tooling
Aug 2020 to present
Technical Creator
TheUnityGuy
Created more than 60 tutorials, video essays and development vlogs, reaching more than 400k views and 6k subscribers, with brand collaborations including Brilliant and Meshy.
Location
Remote
60+ videos450k+ views6k+ subscribers

Toolkit

The tools I actually use.

I've picked up some from school, hackathons, personal and commercial projects, and proffessional experiences.

Tool Map

Grouped by how I typically use these frameworks/tools.

0 tools

Languages

9 tools

TypeScript

JavaScript

Python

Java

SQL

C++

Backend

15 tools

Node.js

FastAPI

GraphQL

WebSockets

PostgreSQL

Prisma

Supabase

Redis

Kafka

Docker

Kubernetes

AWS

GitHub Actions

Linux

Git

Frontend

5 tools

React

Next.js

Tailwind CSS

HTML5 Canvas

Leaflet

AI / ML

10 tools

Google Gemini

Anthropic

Cerebras

IBM watsonx

ElevenLabs

Whisper

Composio

Backboard

Pillow

mssmss

Creative / Game Dev

4 tools

Unity

Blender

Adobe Photoshop

YouTube

Hackathons

The ones that placed.

Weekends that ended with a sneaky win.

Track winnerJuly 2026
Hack the 6ix
Hack the 6ix 2026 · Most Innovative Use of Backboard.io
Recompile — A wellness companion that reads your vitals and your commits, then waits for a boundary to interrupt.
1st placeMarch 2026
GDG Build with AI
Google Developer Group · Case competition
Flux — An AI study system that turns lectures, PDFs, and notes into podcasts, games, flashcards, and quizzes.
LinkedIn GitHub
2nd place2026
Cursor Hackathon
Toronto Tech Week
Haggle — Agents that find sellers, call them by phone, and negotiate a better price.
LinkedIn GitHub
1st placeJuly 2026
United Hacks V7
Online hackathon
PitchSide — Real football matches rebuilt from their recorded event streams into navigable 3D worlds.
GitHub
3rd place2026
Agent Shyft Hackathon
Toronto Tech Week
Synapse — A classroom platform where uploaded course material becomes adaptive study modes.
LinkedIn
Track winner2026
Sand Castle Hackathon
Toronto Tech Week · Know Idea track
Tide & Till — A cozy fishing game built on real marine-life and recipe data.
GitHub Devpost