• 

  Logo

• 

  Homepage dark mode

• 

  Homepage light mode

💡 Inspiration

We wanted to bridge the gap between search engines and AI assistants. Traditional search engines provide links without context, while chatbots generate answers without verifying sources. The idea behind SearchMate was simple but bold — create a trustworthy, AI-powered search tool that combines both worlds, entirely in the browser, with no backend and no data tracking.

⚙️ What it does

SearchMate is a zero-backend AI search assistant that:

• Retrieves information from Elastic Cloud (hybrid keyword + vector search).
• Summarizes and reasons with Google Gemini (Vertex AI) to deliver concise, cited answers.
• Displays results in a minimalist, fast, privacy-friendly web interface.

In short: Ask anything. Get a clean, verified answer — instantly — with real sources.

🧩 How we built it

We built SearchMate using a frontend-only stack:

• Next.js 14 + TailwindCSS for a modern and responsive UI.
• Elastic Cloud API for fast hybrid search on indexed documents.
• Google Gemini API for natural language understanding and generation.
• Vercel for one-click hosting and deployment.

All requests are made directly from the client using read-only, referer-restricted API keys — ensuring zero backend maintenance and total scalability.

Architecture overview:

User ─▶ Browser (Next.js)
       ├── Elastic Cloud → retrieves relevant passages
       └── Gemini API → synthesizes contextual, source-cited answers

🧗‍♂️ Challenges we ran into

• Handling CORS and API key security without a dedicated backend.
• Optimizing prompt design to ensure Gemini gives short, accurate, and source-cited answers.
• Designing a fast, elegant UI that feels like a native app with no page reloads.
• Managing token limits when sending multiple document contexts to the AI.
• Ensuring the experience works seamlessly across browsers and devices.

🏆 Accomplishments that we're proud of

• Built a fully functional AI RAG system that runs entirely client-side.
• Achieved zero infrastructure cost — deployable in minutes with no servers.
• Created a privacy-first search assistant: nothing is stored, everything runs locally.
• Designed a clean, modern UI inspired by Perplexity and Gemini.
• Successfully combined Elastic + Gemini in one unified experience.

🧠 What we learned

• How to leverage Elastic hybrid search for real-time context retrieval.
• The importance of prompt engineering when chaining retrieval + generation.
• Using referer-based security to safely call APIs from the browser.
• How minimal frontends can still feel powerful, intelligent, and secure.
• That innovation doesn’t always require complexity — sometimes simplicity is the real edge.

🔮 What's next for SearchMate

• Add Google Drive, Gmail, and PDF indexing for personal knowledge search.
• Introduce multilingual support (English, French, Spanish).
• Integrate Gemini embeddings for deeper semantic understanding.
• Develop a mobile PWA version for on-the-go access.
• Open-source the project to empower others to create their own AI-powered search tools.

Built with 💡 by Laurent Goulenok for AI Accelerate (Google Cloud × Elastic).

Built With

• elastic
• elastichosts-cloud-computing
• elasticsearch
• figma
• gemini
• github
• next.js
• tailwindcss
• typescript
• vercel