The Matching Engine

Cosine Similarity Matching#

At its core, OSBridge compares two vectors: your skill profile embedding and each issue's embedding. The closer these vectors are in 768-dimensional space, the better the match.

The Math#

Cosine similarity measures the angle between two vectors. A value of 1 means they point in exactly the same direction (perfect match), 0 means they're perpendicular (no relation), and -1 means they're opposite.

-- pgvector cosine distance query
SELECT 
  id,
  title,
  1 - (issue_embedding <=> skill_embedding) AS match_score
FROM issues
WHERE 
  is_open = true
  AND has_linked_pr = false
  AND comment_count < 10
ORDER BY issue_embedding <=> skill_embedding
LIMIT 20;

Filtering#

Not every issue makes it through. We apply strict freshness filters to ensure you only see actionable issues:

• Must be open (no closed issues)
• No linked pull requests (someone else isn't already working on it)
• Comment count under 10 (lower competition, more welcoming)
• Minimum match score threshold of 40%
• Issue not older than 90 days

Ranking#

After filtering, the top 20 issues are ranked by match score. The top 5 receive a custom AI-generated "Why You" explanation, making it clear why your specific skills qualify you for the issue.