GenAIProduction

MocktailVerse Bedrock

Ask for a vibe, not an ingredient — AWS Bedrock RAG finds the drink that matches what you meant.

🧠 AWS

Bedrock Claude + Titan

🔍 pgvector

Postgres vector search

🔁 RAG

Chunk → retrieve → answer

🏭 E2E

Ingest to generation

Tech Stack

PythonAWS BedrockpgvectorRAGPostgreSQLLangChainboto3DockerFastAPI

Before

Keyword cocktail search

Question

↓

SQL LIKE search

↓

Exact-name match

↓

Misses the vibe

Ask for 'something smoky and bitter' and a keyword search shrugs.

Needs the exact ingredient or drink name to find anything.

No semantic understanding — it matches letters, not meaning.

Real vector search usually means a managed vector DB bill.

After

Bedrock RAG platform

Question

↓

Semantic retrieve

↓

Bedrock grounds

↓

Real answer

✓

Vibe SearchAsk for 'smoky and bitter' and pgvector surfaces the mezcal negroni — even when those exact words aren't in the recipe.

✓

Model SwapClaude or Titan behind one Bedrock API. Change the brain without touching the application code.

✓

No-VendorDB VectorsProduction semantic search on plain Postgres + pgvector — no managed vector DB to pay for.

✓

Cost ReceiptsRetry logic, observability hooks, and per-query cost tracking — it behaves like prod, not a notebook.