Lewys

Inspiration

We liked the concept of it, but we realized the problem is closer to us than we expected. Close family that enjoys DIY renovating, the Lowe's representative that talked about how sometimes there just aren't enough associates to help you choose your products, but when they are the conversational aspect makes it feel authentic and meaningful.

We wanted to help create a connection like that between the different kinds of people that reach their home improvement needs through Lowe's.

We don't want people to feel like they're alone, we want them to know they're part of a community that is there to help, so we made Lewy.

What it does

It provides a space for the beginners, the curious, and the experienced to naturally mingle. By having a forum based model, we're able to match the ones willing to learn with the ones willing to help.

Through sentiment analysis and post categorization, tags allow the user to find the posts they're interested in, allowing for more directed interactions.

How we built it

Lewys was built using a React.js frontend and a Firebase backend, connected to a FastAPI-powered LLM for intelligent tagging. With:

Frontend (React.js)
Authentication handling(Google Sign-In)
Displays feed of posts and comments
Allows users to create and interact with posts
Backend (Firebase + FastAPI) to manage users and posts
Firebase handles Authentication and Firestore database storage
FastAPI runs an LLM that auto-tags posts using natural language understanding
ngrok provides a secure development tunnel to the backend

Challenges we ran into

LLM + FastAPI: Running a local model off of Ollama, and using advanced prompt engineering techniques to derive insights from user tone, wording, and key words was difficult with how variable the model is, especially when it's small.

It was difficult coupling the firebase and python backend with each other. Not everyone was on the same page because of the different skillsets, and we'll only continue to get better at communicating and working with each other going forward.

Getting authentication and data transfer to work smoothly between the two systems required careful setup of credentials and async requests.

Accomplishments that we're proud of

Built a fully functional prototype of Lewys: a collaborative platform connecting homeowners, DIYers, and professionals for home improvement projects
Successfully integrated an AI-powered auto-tagging system using FastAPI and an LLM to categorize posts intelligently by service area and help level
Created a responsive, user-friendly React interface that displays posts, comments, and AI-generated tags in real time.
Implemented secure Google Sign-In and data persistence using Firebase Authentication and Firestore.
Achieved seamless communication between the React frontend, Firebase backend, and FastAPI AI layer through nGrok tunneling.
Overcame major debugging challenges, from CORS issues to async data flow, within a short hackathon timeframe.
Built a project that feels scalable and practical, with real potential to support local communities and sustainable DIY collaboration.

What we learned

Here's a couple of insights we took away:

How to integrate machine learning (LLMs) into a real-world web app using FastAPI and secure API calls.
The importance of state management and real-time data handling in React for dynamic user feeds.
How Firebase Authentication and Firestore can simplify backend logic while keeping data secure and synchronized.
The challenges of frontend-backend interoperability, especially with ngrok and CORS configurations.
Effective team collaboration under time pressure, including clear version control, task division, and debugging together.
How to design software that bridges technical innovation and real human utility — building not just an app, but a helpful community tool.
Making a MVP and prioritizing the main features over others, and debugging in a timely manner

What's next for Lewys

We're very excited to continue working on it. Getting the comment section to work, being able to filter posts based on category, and improving the UI.

And maybe getting user feedback!