Sanjeev Kumar's Projects

Semantic search engine for your documents, using Cohere embeddings

Railway AI Docs Search (Cohere + pgvector)

	Processes 16 chunks per API call; small enough to avoid timeouts, large enough for speed

50 characters overlap prevents cutting important context mid-sentence

~500 characters per chunk; good for semantic search accuracy without excessive token usage

Cohere_Trial_API: https://dashboard.cohere.com/api-keys

Postgres connection string for pgvector (private URL, SSL disabled for internal Railway network)

Private port for Postgres is always the default 5432 inside Railway’s private network