Wizard AI: Building a Real-Time AI Answer Engine with AWS + Google Search

Building Wizard AI: A Smarter, Accountable Answer Engine Using Google Search + AWS AI

Proud to share a project I’ve been hands-on with — Wizard AI – Next GenAI, built at the British Institute of Technology and E-Commerce as part of the Wisdom Program.

What started as an idea in a brainstorm with Professor Farmer is now a live MVP that fuses the power of Google Embedded Search, AWS Bedrock, and custom algorithms — all to give students and researchers a fair shot at quality, reference-backed insights.

What Is Wizard AI?

Not just another chatbot. Wizard AI is a real-time answer engine that pulls fresh, credible data from the web, processes it with a low-cost LLM (via AWS Bedrock), and delivers insightful responses — backed by real sources.

We’re basically trying to build a better version of Perplexity AI, tailored for learning, research, and business needs.

The Algorithm Behind It

We built a simple but effective four-step pipeline:

Step 1: Embed the Query
User inputs a question. That query is converted into an embedded text vector for semantic similarity.

Step 2: Fetch Relevant Docs from Google
We use Google Embedded Search to pull in real-time web content related to the query.

Step 3: Generate Insights
Using a low-cost AWS AI model (via Bedrock), we combine the user query with the meta content from retrieved documents to generate reliable, smart answers with RAG.

Step 4: Show Results
The final output — insights + references — are delivered to the user via a slick frontend powered by React.

🛠️ Stack We Used

• FastAPI (for backend and API gateway)

• ReactJS (frontend interface)

• Google Embedded Search API (for live web search)

• AWS Bedrock (LLM inference on the cheap)

• Custom RAG Flow (Retrieval Augmented Generation to mix Google data + LLM reasoning)

💡 What I Focused On

I was involved in end-to-end dev — from pipeline design to MVP buildout and production deployment. Fast iterations, quick testing, and just shipping it.

The learning curve? Worth it. Especially understanding how to fuse real-time data with LLMs and keep costs low (token Control) without sacrificing quality.

🔄 Still Evolving

Wizard AI is live now — check it out: https://wizard.net.co/

But we’re not done. We're testing new features, improving data quality, and adding AI agents to support more business use cases.

✅ Looking for contributors — if you're into LLMs, RAG, or full-stack AI systems, hit me up.