RETRIEVAL & AI

RAG Knowledge Chatbot

Ask about the work without leaving the portfolio.

01 / THE SYSTEM

The problem.

A visitor may know the question they want to ask without knowing which case study to open. The profile terminal gives them a short path to a relevant answer.

Role I built the knowledge-entry format, retrieval service, prompt assembly and terminal conversation interface.

Approach I connected Ask AI to the portfolio’s retrieval and chat route. Visitors can ask about skills and projects in place. A scope check redirects unrelated questions back to the work.

Next.jsSupabasepgvectorOpenAIVercel AI SDK
Portfolio capture

Ask about the portfolio

Loading imageAsk AI answers a question about the $9 focaccia. Avery and Morgan split it equally in the Finance demo.
A visitor asks who splits the $9 focaccia in the Atomos Finance demo. Ask AI finds the two guests in the dinner walkthrough. Avery and Morgan each take one share.Open full-size still (opens in a new tab)
02 / ARCHITECTURE

How it connects.

Semantic search / reviewed project notes / follow-up questions

  1. 01KNOWLEDGE BASE
  2. 02PGVECTOR SEARCH
  3. 03SOURCED CONTEXT
  4. 04AI RESPONSE
03 / THE DETAILS

Implementation evidence.

01

A question in context

The profile terminal displays the visitor’s question and the reply together. The same panel handles questions outside the portfolio’s scope.

02

Where answers come from

Each search starts with reviewed project and experience notes. If the hosted index is unavailable, the assistant can search the same notes using a local copy of their embeddings.

PROOF

Retrieved text passes through the public naming rules

The assistant searches a reviewed set of public notes. Each match points back to a named project or portfolio section.

Outcomes

  • Editing and re-indexing an entry updates the material available to future answers.
  • The assistant can acknowledge a gap when a visitor asks about an undocumented detail.