README.md

metadata

title: Swift Stock Screener
emoji: 💹
colorFrom: green
colorTo: gray
sdk: gradio
sdk_version: 5.29.0
app_file: app.py
pinned: true
license: apache-2.0
short_description: Search over 12k companies by theme (VSS) or similarity (NNS)

Overview

Swift Stock Screener (SSS) is an interactive application that enables semantic and similarity-based stock screening across ~13,000 equities.
It combines semantic vector search on company descriptions with nearest neighbor search on financial fundamentals.

The app allows two main modes:

• Theme search: find stocks relevant to a concept (e.g., “lithium”, “cloud computing”).
• Ticker similarity search: find peers to a given stock (e.g., NVDA → other semiconductors).

Data (fundamentals and descriptions) is sourced from Yahoo Finance. Company descriptions used in the semantic search come directly from Yahoo Finance long summaries.

---

Features

• Semantic Theme Search
  Uses FinLang/finance-embeddings-investopedia to embed queries and company descriptions.
  DuckDB with HNSW vector index executes approximate nearest neighbor search.
  Re-ranking adds:

  • Literal rewards (companies explicitly mentioning query terms boosted).
  • Brevity penalties (companies with very short descriptions are penalized).

• Ticker Similarity Search
  Uses normalized numeric fundamentals (e.g., Market Cap, P/E, Revenue Growth, Volatility).
  Scikit-learn NearestNeighbors finds the closest peers.
  A Similarity Index (0–100) is computed with a gamma distribution scaling to give intuitive scores.

• Interactive Results Table
  Paginated grid view with columns:

  • Name, Country, Sector, Market Cap
  • 1yr Return, Volatility, P/E, Revenue Growth, Dividend Yield, Beta
    Negative values are highlighted red.

• Dynamic Filtering
  Click on a sector, industry, or country in the grid to filter results interactively.
  Multiple filters can be applied in succession.

• Company Details View

  • Company name + Yahoo Finance description.
  • Full fundamentals table.
  • Radar chart of 5 normalized metrics: Beta, Debt/Equity, 1yr Return, Revenue Growth, Volatility.

• Controls

  • Theme box: free text search.
  • Ticker box: find peers to a stock.
  • Search / Reset / Random ticker buttons.
  • Sorting by any metric.

---

Architecture

• Data: Yahoo Finance fundamentals + descriptions.
• Embeddings: FinLang/finance-embeddings-investopedia (768-dim vectors).
• Vector DB: DuckDB with VSS extension + HNSW cosine index.
• Ticker similarity: Scikit-learn NearestNeighbors with Euclidean distance over normalized features.
• UI: Python + Gradio Blocks.
• Visualization: Plotly for radar chart.

---