AI-Native Portfolio

Overview

A personal portfolio and blog site designed with AI as a core architectural principle. Instead of a traditional static site, this portfolio uses semantic search and RAG-powered chat to help visitors discover and explore content.

Architecture

The site is built with Next.js 15 using the App Router. Blog posts and project pages are written in MDX and statically generated at build time. AI features run through Next.js API routes:

• /api/search — Embeds the query via Bedrock Cohere, then queries S3 Vectors for similar content
• /api/chat — Retrieves relevant context from S3 Vectors, then streams a response via Bedrock Claude

Key Decisions

Why S3 Vectors over Pinecone/Weaviate? Cost (90% cheaper), AWS-native integration, and the durability guarantees of S3.

Why custom RAG over Bedrock Knowledge Bases? Full control over chunking, ranking, and prompt engineering. Also a better learning experience.

Why Vercel AI SDK with Amplify hosting? The AI SDK is the best React chat library available. Amplify handles Next.js deployment natively.

What I Learned

Building AI features for a content site is more about information retrieval design than model selection. The quality of search results depends on chunking strategy and metadata filtering far more than which embedding model you use.