Overview

This project integrates a custom AI assistant into my portfolio, giving visitors an interactive way to learn about my work. By leveraging retrieval-augmented generation and a serverless architecture, I built a responsive chatbot that answers questions from my actual portfolio content while running on free-tier infrastructure.

Key features

• Vector-based knowledge retrieval using Pinecone for accurate information matching
• Natural language processing powered by Cohere’s command-r model
• Serverless implementation with Netlify Functions for efficient, cost-effective deployment
• Responsive UI in two flavors: dedicated chat page and floating widget

Technical details

Vector database

• Pinecone for efficient semantic search
• Embeddings generated via Cohere’s embed-english-v3.0 model
• Cosine similarity matching to find relevant information for each query

Serverless architecture

• Netlify Functions for backend processing and API integration
• Stateless design with conversation history maintained client-side
• Health-check endpoint so future breakage surfaces immediately

How it works

1. User input is converted to an embedding vector via Cohere’s embedding API
2. The embedding is used to query Pinecone for the most relevant portfolio chunks
3. Retrieved context is passed to Cohere’s chat completion API along with the query
4. The assistant is instructed to ground every claim in the retrieved context
5. The response streams back to the UI

Try the chat