Relay-app

Upload Document page
Document uploaded confirmed
Dashboard page
AI chat box

Inspiration

Enterprise knowledge is constantly changing. Important decisions, policies, architecture choices, and tasks are spread across documents and often become difficult to find or contradict newer information.

We wanted to build a system where organizational knowledge is treated as living memory, rather than simply stored documents.

This inspired Relay, an AI-powered Enterprise Memory System that helps teams capture knowledge, retrieve trusted information, detect conflicting decisions, and turn AI insights into actionable workflows.

What it does

Relay provides a centralized AI memory layer for enterprise knowledge.

Document ingestion — Upload enterprise documents and process their contents.
AI memory extraction — Extract important decisions, tasks, and knowledge from documents.
Semantic search — Ask natural-language questions about organizational knowledge.
Hybrid retrieval — Combine vector similarity search with structured database queries.
Grounded answers — Return AI-generated answers with citations to the underlying documents and memories.
Conflict detection — Identify contradictory decisions or policies.
Conflict resolution — Allow users to keep an old decision, accept a new decision, or merge them.
AI-proposed actions — Allow the AI to propose actions such as creating tasks or updating decisions, while keeping the user in control.
Auditability — Track important actions and changes to organizational memory.

How we built it

Relay uses CockroachDB as its persistent enterprise memory layer and *AWS * for AI and cloud infrastructure.

The core architecture is:

User
  ↓
Relay Frontend
  ↓
Backend API
  ↓
AI Agent 
  ↓
Hybrid Retrieval
  ├── CockroachDB Vector Search
  └── Structured SQL Queries
  ↓
Relevant Enterprise Memory
  ↓
AI Reasoning
  ↓
Answer + Citations + Proposed Actions