How we turned Slack chaos into a living Knowledge Graph using LLMs, decay scoring, and graph theory:
The $31 Billion Problem Nobody Talks About
Fortune 500 companies lose an estimated $31.5 billion annually from employees failing to share knowledge. But it's not that people won't share, it's that knowledge gets shared in Slack threads at 2 AM during an incident, in a standup nobody recorded, in a PR review that's now archived.
We built ECHO to fix this.
What is ECHO?
ECHO is an AI-powered Organizational Knowledge Engine. It passively monitors your Slack workspace, extracts structured intelligence from natural conversations, and builds a real-time Knowledge Graph that maps:
Who knows what
When decisions were made and why
Which knowledge is decaying (hasn't been referenced recently)
Who to ask when you're stuck on a specific technology
How It Works
1. Passive Ingestion ECHO connects to Slack via Bolt SDK in Socket Mode. Every message flows through our ingestion pipeline — no buttons to click, no forms to fill.
2. AI Entity Extraction Each message is processed by Groq's Qwen 3 32B model, which extracts structured entities: technologies mentioned, decisions made, problems raised, and people involved. This classification happens in under 200ms per message.
3. Knowledge Graph Construction Extracted entities become nodes. Relationships become edges. A person who mentions "Redis" and "caching" creates person → knows → concept edges. Two people discussing the same topic creates collaborated_with edges. The graph grows organically.
4. Temporal Decay Not all knowledge is equally fresh. ECHO applies exponential decay scoring, knowledge that hasn't been referenced in 30 days fades, while recently discussed topics pulse brightly. This ensures the graph reflects current team expertise, not historical noise.
5. Interactive Dashboard A React + Vite frontend renders the Knowledge Graph as a force-directed visualization with real-time WebSocket updates. Nodes pulse when new messages are ingested. The Expert Finder, Ask ECHO, and Memory Timeline provide three distinct lenses into your team's collective brain.
Architecture
Slack → Bolt SDK → Ingestion Pipeline → Groq LLM → Memory Store + Knowledge Graph → Express API → React Dashboard
The entire system is a TypeScript monorepo with shared packages for types, memory, graph, and agent logic.
Try It Yourself
ECHO is fully open-source. Clone the repo, add your Slack bot token and Groq API key, run pnpm dev:all, and watch your team's knowledge graph come alive.
Top comments (0)