BLUE SKY POTTERY AI EMAIL CHATBOT

PROJECT 02

Pottery Studio Email Chatbot

A full-stack AI email assistant that reads, understands, and drafts replies to customer inquiries for a local pottery studio — built with a FastAPI backend and React frontend.

FastAPI · React · Groq · Email Automation · REST API · Full-Stack

The Problem

The pottery studio's inbox was a bottleneck. Customers emailing about class schedules, pricing, availability, and custom orders were waiting too long for replies because the owner was handling everything manually. Many inquiries were repetitive and followed predictable patterns — exactly the kind of work an AI can handle well.

The studio needed something that felt personal and on-brand, not a generic autoresponder. Replies needed to sound like the studio owner, cover the right information, and give the owner easy control over what went out.

The Solution

I'm building a full-stack web application where the studio owner can see incoming customer emails, review AI-drafted replies, edit them if needed, and send — all from a clean interface. The AI layer reads the incoming email, pulls context about the studio's offerings and policies, and generates a reply in the studio's voice using Groq's inference API.

The FastAPI backend handles email parsing, AI prompt construction, and response generation. The React frontend gives the owner a simple inbox-style interface. Nothing sends automatically — the owner always reviews before anything goes out, keeping them in full control.

ARCHITECTURE DIAGRAM

Key Features

AI-generated email drafts that match the studio's tone and include accurate information about classes and pricing

Human-in-the-loop design — the owner always reviews and approves before anything sends

Clean inbox UI showing incoming emails, AI draft, and edit controls side by side

FastAPI backend with clean REST endpoints separating concerns between email logic and AI logic

Zero-cost infrastructure using Groq free tier and self-hosted backend

Challenges & Solutions

Generating replies that sound human and on-brand

Generic AI email replies feel robotic and impersonal, which would hurt the studio's brand. The solution was careful prompt engineering: the system prompt includes the studio's tone guidelines, sample phrases the owner uses, and structured information about their offerings. The model is instructed to be warm and conversational, not formal or corporate.

Building a full-stack app as a one-person team

Managing both a FastAPI backend and a React frontend simultaneously required careful API design upfront. I defined the request/response contracts before writing either side, which meant the frontend and backend could be developed in parallel without constant integration bugs.

Keeping the owner in control

The temptation with automation is to make it fully automatic. But for a small business where every customer relationship matters, a bad AI reply is worse than a slow human one. Designing the human-review step as a non-negotiable part of the workflow — not an optional override — was a deliberate product decision.

Status & Next Steps

The project is in active development. The FastAPI backend and core email parsing are complete. The React frontend inbox UI is in progress. Next steps include integrating the full Groq pipeline into the frontend, adding email send functionality, and deploying both services for the client.

What I'm Learning

This project is pushing me into full-stack territory in a way the Braxton chatbot didn't. Managing the contract between a Python backend and a JavaScript frontend — keeping data shapes consistent, handling async errors gracefully, designing REST endpoints that are easy to consume — is a different kind of engineering discipline than building a single-service Python app.

The prompt engineering challenge here is also more interesting than Braxton: writing a system prompt that consistently generates replies that feel human requires a lot of iteration, evaluation, and refinement. It's closer to product design than engineering.