SalineGuardAI Agent

Inspiration

I'm not AI developer, i build AI agent because it'ss my the favorite, addicted to. It started with the quiet pain of a friend’s family in the Mekong Delta.Every dry season, they watched the water turn salty. Rice fields that once fed generations turned white and brittle.

He said, “One day, you’ll leave this world behind So live a life you will remember.”

I sing a line from a song that has followed me in during the LotusHack 2026. Every day that passes is a day we can never get back. If today we don’t start doing something or anything, that benefits society, that helps our country, then perhaps the greatest regret we’ll carry tomorrow won’t be that we weren’t rich enough, or famous enough… It will be that we didn’t live a life worth remembering.

What it does

SalineGuard AI Agent is an intelligent conversational AI assistant that helps Mekong Delta farmers fight saline intrusion. Farmers ask natural questions in Vietnamese or English, such as:“Current salinity in Ca Mau today?” “Advice for my shrimp pond at 5.2 ppt?”

The Agent instantly responds with: Real-time salinity data from local stations
Insights from NASA SMAP (sea surface salinity) and NASA POWER (climate/rainfall)
Practical farming recommendations (when to take in water, pump out, harvest early, etc.)
High-risk alerts and trends visualization

It’s built to be simple, bilingual, and impactful, turning complex data into clear decisions for real people.

How we built it

Create a project plan, brainstorm, research AI docs, system prompt, prompt project, secure Data, structure project... Tired, no sleep, low health, low food, ex-team breakdown. overload!! Frontend: React + TypeScript + Tailwind CSS + Vite (generated & refined via Lovable.dev)

Backend & Database: Supabase (PostgreSQL for salinity data, Edge Functions for secure API calls, Realtime for live updates)

Map Visualization: React-Leaflet Rapid Prototyping: Entire app built and iterated in days using Lovable.dev + manual layout fixes. vibe to build AI

Challenges we ran into

First time to use Lovable, no credit ( cannot claim), low loading, error coding... Fixing responsive layout issuses (map overlapping chat on mobile/desktop) Keeping the app lightweight and error-free for live demo Making the Agent truly context-aware and citation (required careful tool schema & system prompt engineering), Data base Build Agent( reasoning, trust, verify) *, not Bot Chat tool.

Accomplishments that we're proud of

Built a real AI Agent (not just a chatbot) that combines local data + NASA satellites to give actionable farming advice
Created a bilingual, farmer-friendly interface with clean dashboard, map visualization, and high-risk alerts
Aligned perfectly with Social & Mobility by TASCO track: meaningful environmental & community impact
Delivered a polished, demo-ready product in a very short time — from idea to live app at https://salinity-guard-vietnam.lovable.app/

What we learned

Prompt engineering is key to making AI Agents reliable and safe (function calling + clear tool descriptions matter a lot). Dont focus to build multi-agents, big system. Real impact comes from simplicity: farmers need natural language, not complex dashboards
Combining satellite data (NASA) with local sources (Supabase) creates much stronger insights than either alone
Rapid iteration with Lovable.dev + Supabase is incredibly powerful for lotushack-speed development

What's next for SalineGuardAI Agent

Integrate official Vietnamese data sources (NCHMF, Sở NN&PTNT, MRC, SERVIR-Mekong)
Add IoT sensor support for real-time field monitoring
Implement push notifications & SMS alerts for high-salinity warnings
Expand to full early-warning system across the Mekong Delta
Open-source parts of the stack and partner with local agriculture organizations

Built With

• javascript
• nasa
• openai
• react
• sql
• tailwind
• typescript
• vite