Runway

Inspiration

From first-hand experience working for a corporation, employees do not receive the proper attention to the feedback they have for the business. Especially when an appliance is broken and needs permission from 3 different layers of "I'll talk to my boss". This application helps solve that by going to the executive in charge and directly requesting money.

What it does

An employee describes an item, quantity, budget, deadline, and required specifications. Runway clarifies missing details and searches for the product; it compares compliant offers by cost, delivery, and specification matches. The employee submits an evidence-based proposal with a link to the source, and the executive reviews it; an approved purchase can be seen in a receipt. These purchases can be tracked from the executive's account and keep in mind of the company's budget and obligations it has such as rent.

How we built it

Frontend: Next.js, React, TypeScript, and Tailwind CSS; hosted on Vercel. Backend: Node.js, Fastify, TypeScript, SQLite and Zod. Gemini: interprets requests and supports research. SerpApi: Google Shopping and product-detail research. Nessie: company budget context, merchant mapping, and simulated purchase records. Reusable agent configurations: shared tools, instructions, and execution limits. Clerk: Google sign-in and employee/executive access. Keep calculations, eligibility checks, and purchasing permissions in backend code.

Challenges we ran into

We ran into a lack of teammates when 2 teammates couldn't make it at the last second, so we split the entire frontend and backend between the two of us. Connecting authorization between the hosted frontend and the backend across domains. We encountered API rate limits during research, so we reduced unnecessary Gemini calls, reused cached SerpApi results, and moved calculations and eligibility checks into backend code.

Accomplishments that we're proud of

Turning a plain-language purchasing need into a structured proposal. Showing evidence and uncertainty instead of presenting every result as a confident recommendation. Connecting research, executive review, and sandbox banking in one workflow. Building a mobile-friendly experience as a two-person team.

What we learned

AI is useful for interpreting needs and explaining trade-offs; code should enforce budgets, permissions, and calculations. The cheapest offer isn’t always the best eligible offer. Clear requirements and shared frontend/backend contracts prevent integration problems. A dependable, focused demo matters more than a long feature list.

What's next for Runway

We plan to scale this project to more sectors so that it covers more purchasing categories and supplier coverage. Add historical spending insights that could be used by executives to analyze spending patterns. Add company-specific purchasing policies.

Built With

• capital-one
• claude
• clerk
• docker
• eve
• fastify
• google-gemini
• javascript
• next.js
• node.js
• python
• react
• serpapi
• sqlite
• tailwind
• typescript
• vercel
• zod