Project · Art Meets Algorithm

HeatSafe

Hands full, sun overhead, question urgent. So talk to it.

A supervisor on a Dubai site at 45 degrees, wind picking up, scaffold crew already halfway up. The rule that decides whether work stops is written down. It is in a PDF, on a laptop, in an office, three floors and one locked door away.

A safety document fails the moment someone needs both hands.

HeatSafe answers out loud. Wind speed, heat, the company procedure, the regional regulation, the manufacturer's guidance. It knows which of those outranks the others, so it quotes your own SOP before it quotes the internet, and it says where the answer came from.

Built in one day at the Dubai AI Hub Builder Lab #3 with ElevenLabs, Context.dev and Devin. Team of five. I was the one who could not code, which turned out to be the useful seat: I wrote the brief, shaped the source hierarchy, and ran the demo.

1day, start to demo
5on the team
3tools in the stack
FinalistBuilder Lab #3
Dubai AI Hub Builder Lab #3, August 2026.

How it works

How HeatSafe answersA spoken question is checked against company rules, regional law, manufacturer limits and live conditions, resolved by authority, and returned as an answer with its source or an honest refusal. A question, out loudhands stay on the work Company SOP Regional law Manufacturer Live weather Authority resolvedstrictest binding rule wins, conflicts are shown Answer with its sourcefact, rule, what it means Or a refusalno source, no guess The human decidesaccountability never moves

How it decides what to trust

  1. Your company's own SOPs, site rules and safety manuals come first.
  2. Then UAE federal and emirate-level requirements, and official HSE guidance.
  3. Then the manufacturer, where the equipment has its own limits.
  4. Live weather and wind data on top, because the rule is useless without today's conditions.
  5. It names its source every time, so a supervisor can defend the call afterwards.
Read the business case See the pitch deck See the build ↗

Built with ElevenLabs for voice, Context.dev for live data and Devin for the code. I directed the brief and the demo. Questions or ideas: say hi.

← Back to ankitabiswas.com