Was unable to present, was not finished in time.
Inspiration
I was inspired by the virtual pet phenomenon (Tomagatchi in particular) and the sort of "jokingly threaten to motivate" campaign Duolingo used. I wanted to build something that used AI on the device (though ultimately I had to abandon this and pivot to OpenRouter's API), and could benefit social good in some way. Procrastination is an issue I run into frequently, so I figured this would be a good subject to tackle.
What it does
In short, it gives you a bunch of virtual pets to care for called "Tasklings," which all rely on you completing tasks in order to be fed/nourished/enriched. They are deliberately desinged to be as cute, pathetic and sympathetic as possible to serve as additional motivation for the user, especially since it is possible for them to die if not properly cared for. AI is used to give the ability to communicate with the Tasklings in more depth to help deepen the bond, and is also used as a sort of "judge" and "life coach" to ensure that the tasks you submit to it are worthy of getting points and are actually being accomplished (thus rewarding actual work and habit change, rather than cheating.) The Tasklings are able to "text" and "call" the user (and leave voicemails) to remind them to care for them (in a way to cause guilt and thus motivate the user to resolve their personal matter and submit proof to the llm.)
How we built it
This project was created using the heavy usage of AI tools (ie, "vibe-coded") (Google Antigravity, with the Gemini 3.1 Pro, Gemini 3 Flash, and Opus 4.6 model backends; Gemini CLI, with the Gemini 3.1 Pro and Gemini 3 Flash backend; Cursor, with the Opus 4.6 and Auto model routing backends; Codex, with the GPT 5.4 Pro backend. All were obtained using the free tier and/or free tier with higher rate limits for students.)
Challenges we ran into
"Vibe coding" is unstable and thus there were significant difficulties in ensuring the product was stable and deliverable on time. Many bugs occurred, the codebase had to be scrapped and restarted based on a reduced specification.
Accomplishments that we're proud of
Coming up with the concept was a major breakthrough for me, I truly think this could be helpful to get people motivated to complete their work if they procrastinate. (Although the psychological impacts are maybe not worth it in the current state, if the user overly bonds with their pets and is yet unable to help them.)
What We Learned
I need to get better at decomposing the tasks into actionable smaller deliverables. I got carried away with scope creep, and was too dead set on getting the local AI integration working to pivot to a strategy that probably would've worked.
What's next for Tasklings - AI Integrated Pet System To Keep You On Task!
Probably a major code refactoring. I tried very hard to get local LLM inference working for this but just ran into too many issues and had to resort to OpenRouter's API to get a passable MVP. That's my next goal. Ideally I'd want to be using Gemma 4 E4B or E2B since it has a live response mode.
Log in or sign up for Devpost to join the conversation.