Inspiration
At a live festival, the challenge is not simply receiving information. It is keeping track of what still matters while everything changes at once. A safety lead like Mo may be listening to radio traffic, coordinating volunteers, tracking incidents, watching for staffing gaps, and making safety decisions while moving around the site. That led us to one central question: What if Mo did not have to constantly monitor everything, but could trust the system to bring his attention to the moments that actually require his judgement?
Ground Control was built around that idea. Instead of giving Mo another dashboard to watch or chatbot to actively question, we wanted the system to work quietly in the background and only bring him in when human judgement is needed. Our guiding principle became: The AI listens so Mo does not have to remember. Mo decides so the AI never has to.
What it does
Ground Control is an AI-assisted safety and volunteer coordination system for live events. As radio calls arrive, it works out:what happened ;where it happened; how urgent it is; which team is responsible; and whether the message is a new incident, an update, or potentially the same incident being reported differently.
Incidents remain open until progress is actually confirmed. If nobody has said they are responding, Ground Control does not assume they are. If nobody has confirmed they are on scene, the incident stays unresolved. When something requires attention, Mo does not receive a full dashboard of everything happening across the festival. He receives one decision at a time.
For example, Ground Control can recommend a responder and show Mo the alternatives, including travel time, suitability, and the operational cost of moving each person away from their current position. The question becomes less about “Who is free?” and more about: “Who can we move without creating another problem elsewhere?” The system also follows incidents through the full lifecycle. Radio updates such as en route, on scene, and resolved update the incident as the real-world situation changes.
Ground Control can also connect differently worded reports that may refer to the same incident and identify wider patterns across multiple calls. Most importantly, it has clear limits. For high-consequence situations such as child safety or red incidents, the system can interpret and recommend, but the final decision remains with a person.
How we built it
Ground Control was built using Next.js, React, TypeScript and Tailwind, with Gemini used to interpret unstructured radio calls and return structured incident information. We separated different kinds of work deliberately. AI handles tasks that require contextual understanding, such as interpreting radio language, linking related reports, identifying patterns, recommending responders, and drafting messages. Ordinary code handles deterministic rules such as timers and escalation thresholds. Human decisions remain mandatory where judgement and safety matter most. We also built a simulated Riverside Festival scenario to test the product under pressure, including staffing shortages, medical incidents, repeated reports, changing responder availability, and unresolved calls. The interface was designed around two different users: Mo, who receives a focused mobile experience with one decision at a time; and the control tent, which has access to the wider operational view.
Challenges we ran into
One of our biggest challenges was deciding where automation should stop. It was easy to keep adding more automated actions, but safety-critical situations forced us to think carefully about which decisions should always remain human. We also had to deal with ambiguity in radio communication. Two people can describe the same incident differently. Two incidents can happen in the same location without being related. A direction mentioned over the radio does not necessarily mean the incident happened there. Another challenge was resisting the temptation to build a traditional dashboard. Dashboards are useful for a control room, but Mo is moving around the festival and may only look at his phone for a few seconds at a time. That pushed us toward a much more focused design: fewer alerts, fewer screens, and clearer decisions.
Accomplishments that we're proud of
We are especially proud that Ground Control does more than classify radio calls. It maintains context across the shift, follows incidents through to resolution, catches potentially repeated reports, identifies emerging patterns, and gives Mo meaningful trade-offs before asking for a decision. We are also proud of the product's quiet state. When Ground Control displays: “No decisions waiting.” that is not an empty screen. It means the system is tracking what is happening in the background and does not currently need Mo's attention. That reflects one of the core design goals of the project: technology should reduce cognitive load rather than create more of it. We are also proud that the system keeps human control visible throughout the experience instead of treating human approval as an afterthought.
What we learned
The biggest lesson we learned is that useful AI does not necessarily mean giving AI more authority. Its value can come from maintaining context, reducing cognitive load, surfacing trade-offs, and knowing when to ask a person for help. We also learned that human-in-the-loop design works best when it is built into the product from the beginning rather than added as a final approval step. Finally, we learned that good operational software is often about removing information, not adding more. For Mo, the goal is not to see everything happening across the festival. The goal is to see exactly what matters when it matters.
What's next for Ground Control
The next step for Ground Control is to move beyond a simulated environment and test the system in real event operations. Our immediate priorities would be: integrating real-time radio transcription;connecting live volunteer rosters, availability and staffing data; improving location tracking so responder recommendations reflect real movement across the site; allowing organisers to configure escalation thresholds and standing orders for different event contexts; strengthening the audit trail so control teams can review what happened, what the AI recommended, and what decisions were made; and testing the system directly with safety leads and volunteer coordinators to understand how it performs under real operational pressure.
Longer term, we see Ground Control as more than a festival tool. The same model could support other fast-moving environments where teams coordinate through radio, information changes constantly, and human judgement still matters. The principle would remain the same: Let the system carry the information, so people can focus on the decisions only they should make.
Built With
- claude-code
- elevenlabs
- gemini
- google-genai-sdk
- next.js
- react
- tailwind
- typescript
- vercel