Empathetic Agents is a one-evening hackathon about building AI that shows up when life gets hard.

Most software treats a person in crisis like a support ticket: a form to fill, a queue to wait in, a script to follow. But the moments that matter most, like a diagnosis, a loss, a layoff, a night you can't sleep, call for something that can actually understand, respond with care, and help you take the next small step, without overstepping what an AI should be doing.

The question for the event: what would it take for an AI agent to be genuinely empathetic and genuinely safe, guiding someone through a tough chapter of life without doing harm?

Hosted by EverSettled, with ElevenLabs and Lovable providing tools and credits for every team, this is a build night for people who want to work on something that actually matters, not just another chatbot demo.

Schedule

• 5:00–5:30 PM — Doors open, food, team formation
• 5:30–5:45 PM — Welcome and challenge briefing
• 5:45–8:30 PM — Hacking
• 8:30–9:15 PM — Demos
• 9:15–9:45 PM — Judging
• 9:45–10:00 PM — Awards and close

In teams of 2-4, you'll prototype an agent against one of two tracks. Both leave room to be creative, one just has a wider brief than the other.

Track 1: Presence

An open, exploratory track. Build an agent that helps someone through something hard, not by answering one question well, but by being there for the stretch of it. Two directions to take it, or blend them:

• A copilot — an agent that helps someone find distraction, meaning, or joy alongside the hard thing they're going through. A bucket list generator for someone with a new diagnosis, a late-night chat companion, a way to plan a good day in the middle of a bad month.
• The infrastructure to track progress and make decisions — the system underneath: what does it remember, and how does it decide when to check in versus when to leave someone alone.

Judged on originality, technical execution, applicability, and how genuinely the agent responds to a person going through something hard.

Track 2: Guardrails

A narrower brief, same freedom to solve it your way. Build the layer that catches an empathetic agent when it gets something wrong.

• AI guardrails for dealing with death — the checks that stop an agent overstepping, sounding falsely certain, or saying the wrong thing around death, grief, legal, or financial matters.
• An outlier detector — a layer that spots when someone's signal is a real crisis, not normal frustration, and routes it differently.

Judged on originality, technical execution, applicability, and how well it actually catches the thing it's meant to catch, based on what you show us in the demo.

Every submission is judged on the same four criteria: originality, technical execution, applicability, and fit to the brief. No pre-built test cases, no hidden benchmark, judges are scoring what you show them live.

1. Register and come with or without a team, we'll help you form one at 5:30 if you don't have one yet.
2. Pick a track at the briefing: Presence or Guardrails. You can mix ideas within a track, you don't have to build exactly one of the four examples above.
3. Use the tools on offer. All participants get free credits for ElevenLabs and Lovable for the night:
   • Lovable — describe your app in plain language and get a working interface fast. Useful if your team has more idea than infrastructure.
   • ElevenLabs — natural, expressive text-to-speech and conversational voice. Useful if your agent needs to actually sound like it's paying attention, not reading a script.
4. Build for 3 hours. You don't need to be an AI expert, bring curiosity, care, and your laptop.
5. Demo it. Show us what it does, not just what it's supposed to do.

Cash prizes across both tracks, plus a month of ElevenLabs Creator Plan for the winning teams. AI credits, pizza, and drinks are on us.

• Originality
  
• Technical Execution
  
• Applicability
  
• Fit to the brief