AI has long since stopped being just about chatting. The real power emerges when an AI model understands and can retrieve your own knowledge, documents, manuals, emails, or project information. That's where embeddings come in.

Gemini Embedding 2: Give Your FileMaker Solution an AI Memory

With the introduction of Gemini Embedding 2, Google has made one of the most powerful embedding models available today. But what exactly are embeddings, why do they matter, and how can you use them within a FileMaker solution?

What are embeddings?

An embedding can be seen as a numerical representation of text.

When you pass a document, manual, or email through an embedding model, it is converted into a long list of numbers (a vector). Those numbers do not represent the exact words, but rather the meaning of the text.

As a result, two texts that share the same meaning can lie close together in what is called the vector space, even if entirely different words are used.

Example

Document

How do I create an invoice in FileMaker?

User question

What is the way to generate a new bill?

To a traditional search engine, these look like two different texts. To an embedding model, they are virtually identical.

This is what makes semantic search possible.

Why Gemini Embedding 2?

Google developed Gemini Embedding 2 as the successor to earlier embedding models.

Key advantages include:

Higher accuracy in semantic search
Strong performance on business documents
Support for multiple languages
Suitable for large knowledge bases
Lower cost than repeatedly sending complete documents to a Large Language Model
Fast processing of large volumes of data

For organisations that want to combine AI with their own knowledge, this is an essential building block.

How does it work?

The process consists of four steps.

Step 1: Gather your knowledge

Think of:

Manuals
Procedures
Project documentation
Client files
Support tickets
Contracts
Product information
Safety documentation

Step 2: Create embeddings

Each piece of text is sent to Gemini Embedding 2.

The model returns a vector such as:

json [ 0.0245, -0.8121, 0.4431, ... ]

This vector is stored in a vector database.

Step 3: User asks a question

For example:

How do I add a user to the system?

This question is also converted into an embedding.

Step 4: Search for similar knowledge

Vector search is used to find which documents are closest to the question.

The retrieved documents are then sent along to a Large Language Model such as Gemini, Claude, or ChatGPT.

We call this process:

Retrieval Augmented Generation (RAG)

With RAG, an AI model can provide answers based on your own knowledge rather than relying solely on its training data.

Why is this interesting for FileMaker?

Many organisations store their business knowledge in FileMaker.

Think of:

CRM data
Tickets
Project files
Knowledge bases
Product documentation
Safety procedures
Contract information
Internal manuals

Traditionally, an employee has to search using keywords.

With embeddings, a user can simply ask a question:

What procedure applies when an employee reports a near-miss?

The AI then automatically searches for the most relevant documents and generates an answer based on the information found.

Practical example: HSE platform

Suppose you manage an HSE platform containing:

Toolbox meetings
Safety instructions
Risk analyses
Incident reports
Procedures
Work permits

By embedding all documents, an intelligent knowledge assistant is created.

Users can ask questions such as:

What should I do in the event of a fall incident?
Which PPE is mandatory for working at height?
What does our procedure say about reporting a near miss?
What training is required for these activities?

The AI searches directly across all available documentation and presents an answer including source references.

Combining Gemini Embedding 2 with FileMaker

A possible architecture looks like this:

text FileMaker │ ├── Documents ├── Tickets ├── Procedures └── CRM data │ ▼ Gemini Embedding 2 │ ▼ Vector Database (Pinecone / Weaviate / Qdrant / pgvector) │ ▼ Gemini / Claude / ChatGPT │ ▼ Answer

New documents can be processed automatically as soon as they are saved in FileMaker.

This keeps the knowledge base up to date at all times.

When do you use embeddings?

Embeddings are valuable when you:

Have more than a few hundred documents
Want to support natural language use
Want AI to answer based on your own knowledge
Want to build intelligent search functionality
Want to feed chatbots with business information
Want to automate support processes
Want to unlock knowledge currently hidden within documents

For simple keyword search, embeddings are often not necessary.

For AI-driven knowledge platforms, they have become virtually indispensable.

Practical applications

Within FileMaker, we are seeing applications like the following increasingly often:

AI Helpdesk

Allow users to ask questions about manuals, procedures, and work instructions.

Smart document search

Find documents by meaning rather than by keywords.

Customer support

Automatically retrieve relevant support tickets from the past.

Safety management

Make safety procedures directly accessible to employees on site.

CRM Intelligence

Let AI gather relevant customer information from thousands of notes and contact moments.

Project assistant

Quickly find previous project experiences, solutions, and documentation.

Conclusion

Gemini Embedding 2 forms an essential building block for modern AI solutions.

By converting business knowledge into vectors, it becomes possible to retrieve information not just by words, but above all by meaning.

For FileMaker developers, this opens the door to:

Intelligent knowledge bases
AI assistants
Semantic search engines
Advanced RAG solutions
Smarter helpdesks
Knowledge-driven chatbots

Where Large Language Models handle the generation of answers, embeddings ensure that the right knowledge is found.

And it is precisely that combination that makes business AI truly valuable.

Want to know more?

Would you like to know how to combine Gemini Embedding 2 with FileMaker, a vector database, and an AI model such as Gemini, Claude, or ChatGPT?

Get in touch with Loggix. We help organisations build AI solutions that leverage their own data, processes, and knowledge.