rag-pgvector

RAG with pgvector: Vector Search in PostgreSQL

No separate vector database needed -- pgvector adds vector similarity search directly to PostgreSQL. This pipeline embeds the query, runs a pgvector similarity query against the specified table, and generates an answer.

Workflow

question, table
       │
       ▼
┌──────────────┐
│ pgvec_embed  │  Embed query
└──────┬───────┘
       ▼
┌──────────────┐
│ pgvec_query  │  pgvector similarity search
└──────┬───────┘
       ▼
┌──────────────┐
│ pgvec_generate│  Generate answer
└──────────────┘

Workers

PgvecEmbedWorker (pgvec_embed) -- Uses Arrays.asList() for the fixed embedding. Calls OpenAI when API key is set.

PgvecQueryWorker (pgvec_query) -- Simulates a pgvector query (SELECT ... ORDER BY embedding <=> query_vec LIMIT K). Returns 3 rows as HashMap entries with content and cosine distance scores.

PgvecGenerateWorker (pgvec_generate) -- Calls gpt-4o-mini with the retrieved rows as context.

Tests

17 tests cover embedding, pgvector query simulation, and generation.

Further Reading

• RUNNING.md -- how to build and run this example