This is a supporting submission inspired by my main hackathon project, “Melly & The AI Agent Squad.” This entry focuses on a specific real-world solution developed as part of the overall system. The full concept, automation logic, and AI assistant framework can be explored in the main submission here.
Inspiration
Safety doesn’t have to be complicated. This compact emergency kit empowers frontline workers — especially fuel station attendants — to respond instantly using printed lanyards, kiosk posters, and auto-generated training prompts. The kit runs with zero tech setup, but behind the scenes, it’s powered by automation: ChatGPT, Framer, ElevenLabs, and multilingual flows via Melly, our AI safety assistant. One flip of the lanyard gives the worker everything they need to Alert. Isolate. Act. — across languages.
What it does
Makes workers reflect — literally — on their personal responsibility for safety
Designed for locker rooms, washrooms, control rooms, kitchens, warehouses
Translated into English, Zulu, Xhosa, and Sotho for inclusivity
Fully printable, poster-ready, and customizable
Can be paired with Melly for auto-generated versions and local slang
How we built it
ChatGPT to write the tagline logic and local-language variations
Ideogram + Canva for poster generation
Aster Lab and image enhancement tools for realism
Framer to display the campaign publicly
Works alongside other Melly campaigns (Spill Kit Training, Awareness Boards, etc.)
Challenges we ran into
Designing something tech-powered that didn’t require any tech to use
Keeping the language simple and multilingual without losing urgency
Balancing printable design with field-tested emergency logic
Getting real emergency protocols translated into steps workers would actually follow
Accomplishments that we're proud of
Built solo ✅ Free tools ✅ No dev team ✅ Execution over excuses This entire project — including the AI agent squad, multilingual safety tools, and automation workflows — was built solo using only free tools, real-world logic, and pure hustle. No dev team. No budget. Just execution.
What we learned
AI agents work best with clear roles. Giving each assistant a specific personality and job (like Trainer Teko for training or Ledger Lana for finances) made task routing and automation more reliable — and more human.
🔗 Workflow design matters more than fancy code. Using structured prompts and memory logic inside GPT-4o allowed us to simulate entire operations without a single API call at first.
🌍 Multilingual AI support is crucial. One-size-fits-all doesn’t work in safety training. We learned how to build AI flows that speak to real frontline workers — in their language, with their slang.
💡 Automation isn’t about replacing people — it’s about removing the noise. By assigning AI agents to handle admin, invoicing, document generation, and training, we free up real people to focus on high-value decisions.
⚡ The faster the tool, the higher the adoption. Everything had to be mobile-ready, offline-accessible, and instant. No logins. No waiting.
🧠 Prompt engineering is a full-time skill. Each AI assistant needed different prompting styles. Melly used empathy + clarity. Ledger Lana was all logic. Trainer Teko had to think like a field instructor.
What's next for Melly- The AI Safety Assistant
Launching OHS Genie, a mobile-first version of Melly for small businesses to automate training, documentation, and safety support
📲 Embedding Melly as a Framer widget + mobile app, so teams can ask safety questions and get instant answers
🧾 Building a full library of auto-generated policies, risk assessments, and training materials
🌐 Offering companies the ability to license their own branded version of Melly & her agent squad
💬 Adding live chat support + dashboard for real-time compliance coaching
👷 Partnering with retail, fuel, and construction companies to train workers using Melly offline
Built With
- agent-roles
- agent-task-routing
- and-other-languages-(used-for-melly-and-assistant-agent-voice-outputs)-?-google-drive-+-docs-?-hosting-and-organizing-downloadable-posters
- chatgpt-(gpt-4o)-?-for-prompt-flows
- cross-checking-automation-logic-??-elevenlabs-?-for-voiceovers-in-english
- database
- document
- lanyard
- logic-design
- merch
- mirror-campaign-visuals-?-runway-+-capcut-?-ai-generated-video-snippets-and-editing-for-the-final-demo-?-gumroad-(planned)-?-storefront-setup-for-downloadable-safety-tools-and-kits-?-notion-+-trello-?-planning
- multilingual-translations-?-gemini-studio-(google-ai)-?-for-expanded-content-ideas
- policies
- risk-assessments-?-framer-+-netlify-?-built-and-deployed-the-interactive-web-version-(mobile-+-desktop-responsive)-?-canva-+-ideogram-+-picsart-+-aster-lab-?-poster
- training-content
- zulu
Log in or sign up for Devpost to join the conversation.