RAG CHATBOT Braxton Animal Works

PROJECT 01

Braxton Assistant

AI-powered customer support chatbot for a local small business, built on a zero-budget RAG architecture and deployed to production.

RAG · LlamaIndex · ChromaDB · Groq · Streamlit · Google Sheets

The Problem

Braxton's team was spending significant time answering the same customer questions repeatedly — about hours, services, pricing, and policies. There was no self-serve way for customers to get answers outside of business hours, and no existing tech team or budget to bring in a solution.

The challenge for Summit was to build something genuinely useful that cost nothing to run, was easy for the client to manage without technical knowledge, and could be deployed and maintained by a single high school student.

The Solution

I designed and built a Retrieval-Augmented Generation (RAG) chatbot that answers customer questions using Braxton's own business documentation as its knowledge base. When a customer asks a question, the system retrieves the most relevant chunks of content from the business's documents and passes them to a language model to generate an accurate, grounded answer.

The entire system runs on free infrastructure. There are no ongoing costs to the client. An admin analytics dashboard gives the business owner visibility into what customers are asking, how the bot is performing, and any questions it couldn't answer — feeding a continuous improvement loop.

ARCHITECTURE DIAGRAM

Key Features

RAG pipeline with document-grounded answers — the bot only responds based on Braxton's actual content, preventing hallucinations

Admin analytics dashboard showing query volume, top questions, and unanswered queries

Google Sheets integration logging every conversation for the client to review without any technical access

Zero-cost infrastructure using entirely free-tier services — no ongoing cost to the client

Streamlit chat UI deployable via Streamlit Community Cloud with a shareable URL

Challenges & Solutions

Making RAG work on zero budget

Most RAG implementations rely on paid embedding APIs (OpenAI, Cohere) and hosted vector databases. I had to find a fully free alternative stack that still produced accurate retrieval. The solution was combining HuggingFace's free sentence-transformer models with a locally hosted ChromaDB instance — which worked reliably and with latency acceptable for a chat interface.

Building for a non-technical client

The client needed to be able to see what the chatbot was doing without touching any code. The Google Sheets logging and Streamlit admin dashboard were designed specifically around this constraint — everything the client needs to know is in a spreadsheet they already understand.

Deployment with no DevOps

Getting a Streamlit app reliably accessible to customers without a server or domain required using Streamlit Community Cloud, which provides free hosting for public apps. The deployment process was fully automated so updates push live without manual steps.

Results

Successfully deployed and live — customers can query the assistant 24/7

Client has full visibility into usage through the analytics dashboard and Sheets log

Zero ongoing infrastructure cost to the client or to Summit

Demonstrated that enterprise-grade RAG architecture is achievable on a zero budget

What I Learned

This project pushed me to find creative solutions within hard constraints — no money, no team, a client with no technical background. The biggest insight was that the user experience for the business owner matters just as much as the chatbot itself. Building the admin layer took as much thought as the RAG pipeline, and it's what made the project feel complete rather than like a demo.

I also learned that Groq's free-tier inference is genuinely fast enough for production — response times were consistently under two seconds, which is competitive with paid alternatives.