Matchy Matchup
React Native, Data Science
Brief: iOS and Android app that organises pickleball, tennis and chess sessions. Focus on ease of use and integration with WhatsApp groups.
Challenges: You have 25 tennis players but only enough court space for 16 at a time. They vary in skill level, with some playing for fun and others being more competitive. Players may become frustrated if they have to sit out for too long or if their matchups are significantly stronger or weaker than they are. Additionally, attendance is unpredictable—some players may arrive late or leave early. The session organizer also wants to participate and prefers to spend no more than a minute on admin tasks between rounds.
Matchy Signup
React, fastAPI, python
Brief: A web-based tool for signing up to pickleball, tennis and chess competitions. Focus on ease of use and integration with WhatsApp groups.
Challenges: WhatsApp does not offer built-in functionality for event sign-ups. While many apps and websites allow users to register for events, they typically require account creation, which can be a barrier to participation. Matchy Signup solves this by enabling you to create a secure micro-site that can be linked in the WhatsApp group description. Members can access it without registration, making sign-ups seamless. The site functions like a hybrid between a spreadsheet and a document—easy to edit, allowing users to add their names to the events they plan to attend.
ClingOn
LLM, voice transcription, text embeddings, fastAPI, python
Brief: An AI companion aimed at elders. Integrated with a voice activated intercom for remotely talking to relatives.
Challenges: The main challenge was developing an accessible memory of conversation that could store any variety of knowledge from the names of relatives and medications to opinions on politics and weather. This was achieved by creating a custom text-embedding database and lookup process.
Jolly Roger
LLM, voice transcription, swift, python
Brief: An AI barman who will take your drinks order. Mixed voice transcription and visual interface. Asks for clarification and makes 0 ordering mistakes.
Challenges: Aiming for zero errors is ambitious. Traditionally, this is achieved by repeatedly paraphrasing the user's input and asking for confirmation. Instead, an intuitive system of editable icons was designed to visually represent the customer’s order. These ‘lozenges’ allow for quick corrections with just two taps.
Invisible Stickman
Computer vision, Jetson, TensorRT, python
Brief: Find a socially acceptable way of having cameras in care homes.
Challenges: In old age, people seek dignity and security. Working with investors who owned a group of care homes, we created the 'Invisible Stickman'—a character designed to demonstrate how technology could provide both.