RelayAI – From Idea to Execution

RelayAI is an AI-powered productivity tool that transforms raw ideas into structured execution plans, helping users move from concept to action in seconds.

Comment

💡 What inspired us

The hardest part of building anything—whether it’s a micro SaaS or applying for an opportunity—isn’t the idea.

It’s figuring out how to execute it.

We’ve seen people repeatedly get stuck at the same point: a blank page, with no clear structure on what to do next.

So instead of building another chatbot, we asked:

What if AI could act like a systems architect, not just a responder?

That’s how Relay AI was born.

⚙️ What it does

Relay AI is a Zero-to-One Execution Engine.

You give it a raw idea, and it instantly turns it into a structured, actionable plan.

It supports two modes:

  • Micro SaaS Mode
    Generates a build-ready plan with milestones, architecture, and risks

  • Opportunity Mode
    Breaks down grants, hackathons, or programs into actionable strategies

Instead of copy-pasting responses, users can refine outputs through a live-sync chat that updates the plan in real time.

🛠️ How we built it

We focused heavily on reliability and structure, not just AI output.

  • Frontend: React (Vercel)
  • Backend: Node.js + Express
  • Database: MySQL (structured schema mapping)
  • AI Engine: Google Gemini with strict JSON schema enforcement

We designed the system so AI outputs are predictable, structured, and directly usable, rather than messy text.

🚧 Challenges we ran into

1. Breaking changes in Express (routing crash)
Our backend repeatedly crashed due to stricter route parsing in newer Express versions.
We had to rework routing logic using regex-based handlers to stabilize deployment.

2. AI reliability under real-world limits
Rate limits and timeouts made a single-model system unreliable.
We built a multi-layer fallback system that automatically switches between models when failures occur, ensuring consistent uptime.

3. Long-running AI requests & memory leaks
If a user closed their browser mid-generation, the server kept running.
We solved this using an AbortController to cancel requests instantly and free resources.

🏆 Accomplishments that we're proud of

  • Turning AI output into strict, structured data instead of unstructured text
  • Enabling live updates to plans through chat interaction
  • Building a backend that handles real-world failures gracefully

🧠 What we learned

We learned that building with AI isn’t just about prompts—it’s about systems design.

Handling failures, enforcing structure, and managing long-running processes are what turn an AI demo into a real product.

🚀 What's next for Relay AI

  • Add user accounts and persistent workspaces
  • Export plans into tools like GitHub Issues or Jira
  • Expand into more domains like career planning and business validation

Built With

Share this project:

Updates