Inspiration

My family

What it does

Sombra keeps Central Valley farm crews safe in extreme heat and wildfire smoke. A crew lead describes tomorrow, and six AI agents build a shift plan around the live forecast and California’s heat rules. The plan goes to every worker’s phone in Spanish, Punjabi or English, read aloud. Workers can talk back, and when someone says “me siento mareado,” Sombra gives first-aid guidance and alerts the crew lead right away. Rules engine: Cal/OSHA checks are plain code, so the AI proposes and the rules decide. A plan that breaks a rule can’t ship. Worker phone: Voice input uses Whisper, with server audio for Punjabi, and messages are routed by type (health, request, question). Reliability: Timeouts, key rotation, and a rules-only fallback keep it working if an AI call fails. Stack: [fill in: TypeScript, framework, models, map library, hosting].

How we built it

Tailwind + CSS

Challenges we ran into

Keeping the AI from being the last word on safety, which led us to put the law in code. Getting Punjabi speech and text working end to end. Making the app usable by voice for people who may not want to type. Handling slow or failed AI calls without breaking the plan. Without a native-speaker review of every translation yet, we’re careful about what we claim.

Accomplishments that we're proud of

A usual day plan that the Crítico agent rejects, followed by a revised plan with fewer worker-hours in deadly heat. A rule-breaking plan can’t be sent, no matter what the AI suggests. Play the Day, a 25-second replay from 4am to 7pm with live counters and a real alert to phones. A plan that still works when the AI is down. A project rooted in our own family history: our grandparents were farmers in India.

What we learned

AI is most useful in safety work when it proposes and a deterministic rule engine decides. Designing for the worker first, with large type, voice, and their own language, changes everything else. Trust comes from showing the work: which agent said what, and which rule it cites. Small details like language level, audio and one-tap alerts decide whether a tool gets used in the field.

What's next for Sombra

Native-speaker review of Spanish and Punjabi briefings, and support for more languages, such as Mixteco. Piloting with farm labor contractors and farmworker organizations, and testing on real iPhones and Androids. A wildfire-smoke replay day and a downloadable Cal/OSHA compliance log. Security and legal review, and expiring join codes, before any real deployment.