Inspiration
Planning a trip usually means twenty browser tabs, three spreadsheets, and still guessing wrong on the total. We wanted an agent that could actually reason about a trip the way a well-traveled friend would
What it does
Ameer AI takes a destination, dates, and a budget, and returns a trip plan that actually fits inside it. Underneath the budgeting core, Ameer AI is also a full travel app: users can explore a feed of shared travel experiences, submit their own, and follow other travelers, so the budget plan lives inside a real social product rather than a standalone calculator.
How we built it
The backend is Go with the Gin framework, backed by MongoDB, handling experience posts, user lookup, Google OAuth, and a follow/unfollow system. The frontend is React with TypeScript, with pages for the Explore feed, experience submission, About, Careers, Help Center, and legal pages. Images go through Supabase Storage — the frontend uploads directly and only ever sends the resulting public URL to the backend, keeping the API surface small. We deployed the backend on Render and the frontend on Vercel.
Challenges we ran into
Auth was the recurring fight: cookie domain scoping and refresh-token bugs caused silent logouts, and React's StrictMode double-invoking effects in development masked the real bug for a while. We also had a security scare — a Firebase service account JSON and some GCP API keys got committed to the repo. We had to scrub them from the entire git history with git filter-repo, rotate every exposed credential, and lock down CORS with environment-controlled allowed origins before we felt comfortable deploying publicly.
Accomplishments that we're proud of
Getting from a security incident — leaked credentials in git history — to a fully rotated, CORS-locked, publicly deployed app without losing the timeline. The auth system is genuinely solid now: OAuth, refresh tokens, and cookie handling all work across environments, which is usually where side projects like this quietly break.
What we learned
That the unglamorous plumbing — auth, CORS, credential hygiene — is what actually determines whether an "agent" demo is a toy or a real product. It's tempting to spend all your time on the model's reasoning and none on the infrastructure it sits inside, but a leaked key or a broken cookie domain will sink a launch faster than a mediocre prompt will.
What's next for Ameer AI
Tightening the budgeting agent itself — pulling in live pricing data instead of estimates, letting it re-plan when a user's budget or dates change mid-trip, and surfacing its reasoning ("why this hotel, why this day") directly in the UI instead of just the final numbers.
Log in or sign up for Devpost to join the conversation.