Skill Profiling

From Tags to Vectors#

Your LeetCode profile contains a goldmine of structured skill data. Every problem you solve is tagged with algorithmic concepts like Dynamic Programming, Binary Search, or Graph Theory. OSBridge reads these tags and transforms them into a machine-understandable format.

Step 1: Scraping#

We query LeetCode's unofficial GraphQL API to fetch your solved problem counts by tag. The response looks like this:

// Raw LeetCode tag data
const solvedTags = {
  "Dynamic Programming": 15,
  "Array": 45,
  "Binary Search": 12,
  "Graph": 8,
  "Tree": 22,
  "Hash Table": 31,
  "String": 18,
  "Depth-First Search": 14,
};

Step 2: Normalization#

Tags are normalized to canonical computer science terms. This ensures that variations like "DFS", "Depth-First Search", and "depth first" all map to the same concept.

Step 3: Embedding Generation#

The normalized tags are serialized into a natural language description and sent to Google Gemini's text-embedding-004 model, which returns a 768-dimensional float vector.

// Serialized skill description sent to Gemini
const description = `
  Developer with strong experience in:
  - Dynamic Programming (15 problems solved)
  - Array manipulation (45 problems solved)
  - Binary Search algorithms (12 problems solved)
  - Graph theory and traversal (8 problems solved)
`;