# Building Your Own Knowledge Graph for AI

*Presentation for Ed Cotton's StatMonday Webinar*

🔗 **Follow along:** [howto.fodda.ai](https://howto.fodda.ai)








---

## Today

I'm going to help you make a **knowledge graph-lite** of a set of data that you can query.

**What you'll need:**

- 🛠 Agentic Coder
- 📊 Google Sheets
- 🤖 Anthropic Claude OR Notion
- 📡 Publishing Via Fodda

**Or just follow along.**

---





## Static Reports → Working Knowledge

My business used to be all about trend reports....

🔗 [PSFK Reports on Gumroad](https://psfk.gumroad.com/)

Recently exploring how I can adapt to how people work with co-pilots. So creating read-me and skills files that can be uploaded into Anthropic.

- See newsletter where I post the links to Skills and ReadMe files here: [PSFK Newsletter](https://newsletter.psfk.com/p/eat-shop)
- See the downloads of those files here: [Brand-Owned Dining Skills File Download](https://psfk.gumroad.com/l/Brand-Owned-Dining-as-Retail-Extension?layout=profile)

---







## Providing Expert Knowledge Graphs for AI Systems

Over the least 3-4 months I've been exploring how to provide expert knowledge graphs for AI systems.

🔗 [Fodda Graphs](https://www.fodda.ai/#/graphs)

### Claude Demo

Let me show the Fodda system by demonstrating how it works with Claude.

🔗 [claude.ai](https://www.claude.ai)

The idea behind Fodda - esp with Claude - is that it's  **your data streaming in to your LLM.**

### Try Fodda yourself in Claude by adding one of these MCP URLs as a [Connector](https://claude.ai/customize/connectors)

| Graph | MCP URL |
|-------|---------|
| 🛒 Retail | `https://mcp.fodda.ai/mcp?api_key=sk_trial_retail&user_id=claude-user` |
| 💄 Beauty | `https://mcp.fodda.ai/mcp?api_key=sk_trial_beauty&user_id=claude-user` |
| ⚽ Sports | `https://mcp.fodda.ai/mcp?api_key=sk_trial_sports&user_id=claude-user` |

### Try yourself in Notion by using a [Fodda agent](https://www.notion.so/marketplace/profiles/fodda) and adding one of these MCP URLs 


| Graph | MCP URL |
|-------|---------|
| 🛒 Retail | `https://mcp.fodda.ai/mcp?api_key=sk_trial_retail&user_id=notion-user` |
| 💄 Beauty | `https://mcp.fodda.ai/mcp?api_key=sk_trial_beauty&user_id=notion-user` |
| ⚽ Sports | `https://mcp.fodda.ai/mcp?api_key=sk_trial_sports&user_id=notion-user` |

---




---

## Why A Knowledge Graph?

One way I think about it is that we add instructions within a table or database so that it's easier for an AI to understand the connections between data. E.g. what brand is this data connected to.

- Concepts of knowledge graphs have been around since the 70s, the theory much earlier
- **2012** — Google started to use a Knowledge Graph
- **2025/26** — Seen as a step 'up' from vector databases

For me — as the AI traverses across the data, it promises to uncover connections we would have found hard to find before.


---



## So Let's Get Started — 3 Steps

1. 📊 **The Table** — Set up a Google Sheet to collect raw data and analysis
2. 🔍 **The Analysis** — Use an AI agent to find patterns in your signals
3. 📡 **Publish** — Register on Fodda so AI systems can query your graph

That's it. Fodda handles the MCP connection — you don't need to build a server.





---

## Step 1: The Table

Two tables that work together:
- A table that collects **raw data** — bookmarks, articles, signals
- A table that has an **analysis** of the raw data — patterns, trends

They are interconnected by `pattern_name` — each signal links to a pattern by name.

📋 **Recommended template:** [→ Copy Google Sheets Template](https://docs.google.com/spreadsheets/d/1LqUytJ_Su4CVZvGmxTxDylsQLKmd4yWr5nC8rIhcZAk/copy)

📄 [→ Readme Help File : Setting Up Your Google Sheets Pattern Graph](briefs/brief-sheets-template.md)

📋 **Here's a sheet I updated earlier :** [→ Open Google Sheets Sample]()





## Step 1B: Adding Content to Your Sheet

You can add signals (articles, bookmarks, observations) manually — or you can get an AI agent to build a tool that does it for you.

**Brief 2 does two things:**
1. 📥 **Signal Collection** — Paste a URL, the tool extracts metadata (summary, brands, technologies, sector) via LLM and writes it to your Signals tab
2. 🔍 **Pattern Recognition** — Once you have signals, the tool groups them into patterns (manually or AI-suggested) and writes those to your Patterns tab

**Add this file to a new session in your agentic coder to build the tool:**

📄 [→ Brief 2: Signal Collector + Pattern Recognition](briefs/brief-signal-collector.md)

🔗 **Dashboard:** [howto.fodda.ai/#briefs](https://howto.fodda.ai/#briefs)

👁️ **See Example:** [What Brief 2 builds](https://howto.fodda.ai/example.html)



---

## Step 2: The Analysis (pt 1)

Yes — you could do the pattern recognition yourself.
Or you could use the tool you built in Step 1B — Brief 2 includes an AI-assisted "Suggest Patterns" feature that analyses your untagged signals and proposes groupings.

**If you haven't built the tool yet, add this file to a new session in your agentic coder:**

📄 [→ Brief 2: Signal Collector + Pattern Recognition](briefs/brief-signal-collector.md)

🔗 **Dashboard:** [howto.fodda.ai/#briefs](https://howto.fodda.ai/#briefs)

👁️ **See Example:** [What Brief 2 builds](https://howto.fodda.ai/example.html)





---

## Step 2b: The Analysis (pt 2)

To do the analysis — go to the URL and press **Analyze**.

Do a human review. Return to your agentic coder and suggest ways to run the analysis next time.





---

## Step 3: Publish to Fodda

Your Sheet is your graph. Fodda makes it queryable.

1. Share your Sheet read-only with Fodda's service account
2. Register at [app.fodda.ai](https://app.fodda.ai)
3. Your graph joins the network — discoverable by Claude, Gemini, Cursor, and every MCP-compatible tool

📄 [→ Brief 3: Publishing Your Graph to the Fodda Network](briefs/brief-publish-to-fodda.md)

---

## The MCP URL

I guess for me it's like an **agent talking to an agent**.

Before we had API keys and we had to send very specific information in specific fields. Now it's more like **AI-to-AI chat**: if some information is missing, the AIs will discuss what's missing and go find it (rather than error).

---

## What To Do With Your MCP URL

### Option 1: Claude Connector
You can add it now to Claude as a connector.

🔗 [→ How to set up a Claude Connector](https://app.fodda.ai)

### Option 2: Notion Agent
You can add it now to Notion when you create a new agent.

🔗 [→ How to set up a Notion Agent](https://app.fodda.ai)

### Option 3: Publish & Sell via Fodda
You can register your graph on Fodda to make it discoverable and eventually monetize it.

📄 [→ Brief 3: Publishing Your Graph to the Fodda Network](briefs/brief-publish-to-fodda.md)

---

## For Power Users: Build Your Own MCP Server

If you want full control — custom query logic, caching, or a data source beyond Google Sheets — you can build your own MCP server and register it on Fodda.

📄 [→ Advanced: Standalone MCP Server Brief](briefs/brief-standalone-mcp.md)

---

*Built with [Fodda](https://www.fodda.ai) · [PSFK](https://www.psfk.com)*
