LayoverOS

A context-aware recovery system for stranded travelers with persistent memory, orchestration, and transaction-aware execution.

Open repository

Project claim

Context-aware traveler recovery system with persistent memory.

Role

Agent system architect

Focus signals

agentic commercevector searchpersistent memory

Proof signal 1

Agentic hackathon finalist

Proof signal 2

Persistent memory loop

Proof signal 3

Execution beyond recommendation

Challenge

Travel assistance products feel fragmented unless planning, memory, retrieval, and transaction flow all behave like one coherent system.

Solution

Architected a multi-agent state machine with LangGraph, MongoDB Atlas Vector Search, Coinbase CDP, and Next.js to support memory, zero-hallucination retrieval, and generative transaction UI.

Build notes

Treated memory, retrieval, supervision, and wallet flow as one operating loop.

Used strict retrieval control because trust matters more than clever sounding output in travel planning.

Pushed the experience beyond advice into coordinated execution.

Tools + stack

LangGraph, Atlas Vector Search, FastAPI, and Next.js.

PythonFastAPILangGraphMongoDB AtlasNext.jsCoinbase CDP

Mode

Multi-agent

Memory

Persistent

Transaction layer

Integrated

Architecture flow

Step 1

supervisor

Step 2

agents

Step 3

atlas vector

Step 4

wallet flow

Step 5

ui memory

What makes it special

This project is about orchestration and coherence. The memory, retrieval, wallet flow, and frontend experience are all treated as parts of one operating system.

Why it feels premium

The agent system does not stop at suggestion. It coordinates execution, fallback handling, memory, and transaction UI in one loop.

Decision signals

Orchestrate agents like a system, not a gimmick

Memory should improve execution, not only recall

Tie AI to real product actions

Outcomes

Built a multi-agent state machine instead of a thin assistant wrapper.

Used Atlas Vector Search and strict retrieval control to reduce hallucination risk.

Connected AI workflow to actual transaction and UI flow, which makes the system feel product-grade.