Vibe Search Engine (FAISS + FastAPI)

A local semantic search engine for places, with:

• Sentence-Transformers for embeddings
• FAISS for fast similarity search
• FastAPI for backend
• JavaScript + HTML/CSS for frontend
• Link to copy of Jupyter Notebook: https://colab.research.google.com/drive/1818b_G8VTFzyJA23X3lbZPFQpvIfbPpr?usp=sharing

Acknowledgements

This project was developed as part of the 2025 Corner-DSC-BAC Datathon.

We gratefully acknowledge the support and data provided by our sponsor, corner.

Built collaboratively by:

• Tomas Gutierrez
• Yarden Morad
• Robin Chen

---

Structure

The project should look like this:

• backend/

  • app/
    • __init__.py
    • main.py — App setup
    • core/ — Logic helper functions
      • config.py — Paths/settings
      • embeddings.py — Load model, embeddings
      • faiss_io.py — FAISS indices
      • text_utils.py — Preprocess
    • api/
      • __init__.py
      • routes/
        • search.py — Search

• data/

  • raw/ — Source data, copy from given dataset 2
    • media.csv
    • places.csv
    • reviews.csv
    • AGREEMENTS + LICENSE/
      • data-license.md
      • usage-agreement.md
  • processed/ — Merged data
    • merged.csv
  • indices/ — FAISS indices
    • metadata.index
    • reviews.index

• frontend/

  • index.html
  • static/
    • api.js
    • ui.js
    • style.css

• scripts/

  • build_index.py — Build merged CSV and FAISS indices
  • run_server.sh — Launch FastAPI server

• .gitignore

• README.md

Setup Instructions

Backend Setup

cd backend
python -m venv .venv
source .venv/bin/activate   # or .venv\Scripts\activate on Windows
pip install -r requirements.txt

Data/Training

Extract the given data into data/raw. If data/processed/merged.csv data/indices/*.index doesn't exist, run

python ../scripts/build_index.py

Backend

From the Project root, run

./scripts/run_server.sh
# chmod +x ./scripts/run_server.sh
# ./scripts/run_server.sh

Start the FastAPI server (uvicorn app.main:app) on http://127.0.0.1:8000

Frontend(Static)

cd frontend
python -m http.server 5500

Visit http://127.0.0.1:5500 in browser.

Flow

1. User opens frontend in a browser (http://127.0.0.1:5500)
2. User types query and hit "Search"
3. Frontend sends HTTP request to backend (http://127.0.0.1:8000/search?q=...)
4. Backend:
   • Encodes the query using the model
   • Searches the FAISS index
   • Returns the 5 closest matches
5. Frontend display the matching places