ElderSafe Dining Agent

Inspiration

Families often search for restaurants with elderly parents using very practical needs: no stairs, less slippery floors, paper menus, quiet seating, easy elevators, and clear routes. Normal restaurant search is usually optimized for popularity, distance, and price, but elder-friendly dining decisions are often about risk, accessibility, and confidence.

ElderSafe Dining Agent was inspired by this gap. We wanted to turn a vague family request like “rainy day, taking my dad with weak legs out for Cantonese food, avoid slippery floors” into a safer and more actionable restaurant decision.

What it does

ElderSafe Dining Agent is an AI-powered restaurant search decision layer for elder-friendly dining.

Users can describe their trip in natural language. The agent identifies elder-care constraints such as stairs, slippery floors, paper menus, quiet seating, wheelchair access, elevators, queueing, and whether the store needs phone confirmation before visiting.

It then re-ranks restaurant results into three practical groups: recommended, needs confirmation, and not suitable. Instead of simply showing more filters, it helps families understand which restaurants are safer to try first, which ones require a phone call, and which ones should be avoided.

How we built it

We built the prototype with HTML, CSS, JavaScript, and a Node.js backend. The frontend keeps a familiar restaurant search experience, while the backend provides AI-powered intent understanding and ranking support.

The agent uses an OpenAI-compatible model API layer, designed to work with Qwen / Alibaba Cloud Model Studio through DashScope. The backend parses natural-language dining needs, maps them into elder-care constraints, and supports risk-aware restaurant ranking.

The project is hosted on GitHub, deployed as a public web prototype, and supported by a reusable Codex Skill that documents how to deploy similar Node.js AI prototypes.

Challenges we ran into

The biggest challenge was not adding more AI text, but making the AI useful without making the product feel complicated. Early versions exposed too much reasoning and looked like an internal analysis tool rather than a user-facing product.

We had to simplify the interface, remove repeated labels, reduce unnecessary explanations, and focus the page on the decision a family actually needs to make.

Another challenge was translating vague family concerns into concrete ranking signals, such as slippery floors, stairs, elevator access, paper menus, quiet environments, queueing, and restroom accessibility.

Accomplishments that we're proud of

We are proud that the final product behaves less like a chatbot and more like a search decision layer. It keeps the familiar restaurant search flow, but changes the ranking logic around elder-care needs.

We are also proud that the prototype goes beyond a static demo. It includes a working frontend, a Node.js backend, model API integration, a public deployment path, GitHub source code, and a reusable Codex Skill for future AI prototype deployment.

What we learned

We learned that AI product value often comes from deciding what not to show. Users do not need to see every intermediate inference. They need a trustworthy shortlist, clear risks, and practical next steps.

We also learned that AI agents are most useful when they improve a real decision workflow, not when they simply add a chat box to an existing product.

What's next for ElderSafe Dining Agent

Next steps include connecting richer restaurant and review data, deploying the backend on Alibaba Cloud, using Qwen through DashScope as the primary model provider, adding route and phone confirmation integrations, and testing the experience with real families planning elder-friendly outings.

Built With

• api
• codex
• css
• github
• html
• javascript
• llm
• node.js
• openai-compatible
• render