Labor Locator

Use natural language to search for the services you want in a conversational way!

Comment

Inspiration

Local businesses are highly important to a community, but an issue is that local directories such as Shop Lake Norman have poor search functions, making it difficult for people to find services that they need. Our goal was to enhance the visibility of local businesses while making the search process more user-friendly, especially for those who are less tech-savvy.

What It Does

Our project helps people identify the services they need from the local business directory by allowing for natural language searches. Our GenAI extracts data from business homepages and business cards, using text and image processing to consolidate and present key information in a visually appealing format.

How We Built It

The backend architecture is built on Google Cloud Platform. We use Gen 2 Cloud Functions to run both our text and image Gemini models, Gemini-1.5-Flash-8B, written entirely in Python. We use Google Cloud Storage (GCS) Buckets to store the HTML file of the local business directory, along with configuration files for our chat agent.

The frontend is built using Svelte and JavaScript.

Challenges We Ran Into

One of the biggest challenges was the fine-tuning of our text model. We conducted extensive experimentation with temperature settings and prompt tuning to develop a reliable model that minimizes hallucinations while making reasonable inferences from natural language queries.

Accomplishments That We're Proud Of

  • We were able to use Gemini to set up both text completion and image processing, which was a very fun and rewarding process.
  • We also made a website that we are very proud of, as it is visually pleasing and displays the most important parts of the businesses in a clean and concise way.

What We Learned

  • We learned a lot about developing and configuring GenAI models for different use cases.
  • We also explored various Google Cloud Platform services, such as GCS Buckets, to store key configuration files and HTML. This setup allows us to update information without requiring a full redeployment of our GCP Cloud Function.

What's Next for Labor Locator

  • Pull reviews from online to display as another point of information for businesses.
  • Develop a vector store and custom model to improve information retrieval accuracy while minimizing hallucinations.
  • Implement this on the Lake Norman small business website.
  • Save/log chat history for more conversational flow (question + response with context awareness)
  • Make this a more modular package that can be implemented for any website to add in natural language search
Share this project:

Updates