Future Plans

Future Plans

Vector Search for GitHub Trending

The github_trending table stores repos with summaries and tags as plain PostgreSQL rows. The plan is to add vector embeddings later for semantic search capabilities.

Approach Options

Option A: pgvector on Neon (recommended)

Neon natively supports the pgvector extension. Add an embedding column to the existing table:

CREATE EXTENSION IF NOT EXISTS vector;
ALTER TABLE github_trending ADD COLUMN embedding vector(1536);

Then backfill embeddings from existing summaries and query by similarity:

SELECT repo_name, summary, tags
FROM github_trending
ORDER BY embedding <=> $1  -- cosine distance
LIMIT 10;

Hybrid search — combine vector similarity with tag filtering:

SELECT repo_name, summary
FROM github_trending
WHERE 'ai' = ANY(tags)
ORDER BY embedding <=> $1
LIMIT 10;

Option B: Export to dedicated vector DB

Export existing data and import into a free-tier vector database:

• Pinecone — free tier available
• Qdrant — open-source, free cloud tier
• Weaviate — open-source, free sandbox

Embedding Model Options

Model	Cost	Dimensions
OpenAI text-embedding-3-small	~$0.02/1M tokens	1536
Voyage AI	Free tier available	1024
HuggingFace transformers.js	Free (runs locally)	Varies

Potential Use Cases

• "What trending repos were related to AI this month?" — hybrid search using tags + embeddings
• Trend detection — cluster similar repos to spot emerging patterns
• Personal knowledge base — search your curated trending history naturally
• Weekly digest — aggregate the week's trending repos into a summary report

Other Ideas

• Weekly summary email — aggregate the week's repos into a single report
• Slack integration — send digest to a Slack channel instead of/alongside Telegram
• More languages — add Python, Rust, Go trending pages for broader coverage
• Read tracking — mark repos as bookmarked/starred via Telegram inline buttons
• Star history — track how repos' star counts change over time