Build a Vector Search Cocktail Picker with Claude Code and Momen BaaS

Cici Yu

·June 26, 2026

·2 min read

This project demonstrates a vector search application built during a hackathon with Vibe Coding Collective. The system captures user mood preferences and uses semantic search to recommend exactly one cocktail — eliminating decision paralysis through committed results rather than option lists.

Momen BaaS is a visual, Postgres-native backend that AI coding tools and no-code builders plug into. You configure your data model, AI agents, and logic visually; it exposes a standard GraphQL API your frontend consumes. One backend, many frontends. AI builds your UI; Momen is the backend that survives production.

Live Demo: pick-for-me-mu.vercel.app

One Pick, Not a List

The application intentionally constrains itself to deliver a single recommendation. Key product parameters:

Shared curated catalog of 15 classic cocktails
Mood input via preset energy and vibe selections plus optional freeform text
Two-agent pipeline: query composition followed by vector matching
Persistent storage of one result per user nickname

System Architecture

App Features

Nickname entry with duplicate prevention
Two-step mood quiz with ten preset options each
Optional additional context field
Asynchronous two-agent pipeline delivering one cocktail result with justification

Data Model

Table	Purpose
`energy_option`	10 preset mood intensity options
`going_for_option`	10 preset vibe categories
`cocktail`	15 curated drinks with vector-enabled descriptions
`pick_result`	User selections and results per nickname

Seed data was imported via CSV import functionality.

AI Agents

Agent 1 — Query Composition

Inputs: energy option ID, vibe option ID, optional text
Output: natural language search query

Agent 2 — Pick Selection

Input: search query
Process: Vector similarity matching against cocktail descriptions using cosine distance
Output: Selected cocktail with name, description, tagline, and justification

Vector search is configured in the agent context inside the Momen editor — no separate Pinecone or embedding pipeline required.

Backend Workflows

Two Actionflows handle the logic:

check-nickname-status (sync): Prevents duplicate entries
pick-for-me (async): Executes the two-agent pipeline and persists results

Integration Approach

The backend infrastructure lives in Momen's visual editor while Claude Code generates the frontend. The Momen plugin gives Claude Code direct access to your backend schema — tables, Actionflow inputs/outputs, and AI agent schemas — so the generated frontend code is always correct:

# Claude Code
claude plugin marketplace add momen-tech-org/momen-nocode-plugin
claude plugin install momen-nocode@momen

When backend changes occur, re-introspecting the schema updates GraphQL operations automatically — no manual intervention.

Implementation Timeline

Phase	Time
Backend setup (tables, agents, Actionflows)	~1 hour
Frontend development & deployment	~1 hour
Total	~2 hours

Key Technical Elements

Direct schema access via the Momen plugin — exact field mappings between frontend and database, no guesswork
Embedded vector search — no external vector database required; configured natively in the agent
Sequential multi-agent logic — query composition decoupled from selection
Catalog portability — architecture transfers to any domain with descriptive seed data