Why Prompting is Important

Prompting is the foundation of building an effective Agent. It’s not just about filling fields; it’s about giving clear, actionable instructions that the AI can understand and execute. Here’s why it matters:

1. Clarity Equals Accuracy: A well-defined prompt ensures the AI focuses on the right tasks and produces accurate results.

2. Eliminates Guesswork: The AI relies entirely on your instructions. Poor or vague prompts lead to suboptimal results.

3. Efficiency Through Precision: Detailed prompts reduce errors and save time by minimizing the need for revisions.

4. Experimentation Encouraged: If results aren’t ideal, you can tweak prompts or test with different models in each Axon.

Key Rule: The better you define the instructions, the better your Agent will perform.

What is the Agent Builder?

The Agent Builder is your workspace for designing AI workflows. It’s where you create workflows by combining Axons, which are individual steps, to define triggers, tasks, and actions. Each step is a part of the overall instructions you give to your Agent.

Think of it as assembling a team where each Axon is a member with a specific role—uploading files, processing data, classifying documents, or storing results.

What Are Axons?

Axons are the building blocks of your Agent:

• Trigger Axons: Start the workflow (e.g., file uploads).

• Task Axons: Perform data processing (e.g., extraction, classification).

• Action Axons: Finalize the workflow (e.g., saving or transferring files).

Each Axon has customizable fields where you define its purpose and behaviour.

Step-by-Step Guide: Building a Document Processing Agent

Step 1: Start with the Uploader Axon

The Uploader Axon serves as the trigger for your Agent.

1. Open the Axon Library by clicking the ‘+’ icon in the top-left corner of the Agent Builder.

2. Add the Uploader Axon to your workflow.

3. This Axon allows you to upload documents for the Agent to process.

Why it’s important: Clearly define the trigger to ensure the Agent starts processing files as intended.

Step 2: Add the Classifier Axon

The Classifier Axon helps label and filter documents before further processing.

How to Create an Effective Classifier Prompt

1. Name: This should be clear and specific to the type of document you’re classifying. For example:

• “Employment Contract”

• “Non-Disclosure Agreement (NDA)”

• “Court Case File”

2. Description: This is where you define the condition or question to classify the document. For example:

• “Does this document relate to an employment contract?”

• “Is this document an NDA?”

• “Does this document pertain to a legal case?”

Labels Example

• Label: Employment Contract

Description: “Does this document relate to an employment contract?”

• Label: NDA

Description: “Is this document a non-disclosure agreement?”

• Label: Legal Brief

Description: “Does this document contain legal case information?”

If the document matches the label, it moves to the next step; otherwise, that document will not be processed further.

Why it’s important: A well-structured classifier ensures documents are accurately categorized, enabling precise workflows.

Step 3: Add the Extractor Axon

The Extractor Axon processes files and extracts specific datapoints.

How to Create an Effective Extractor Prompt

1. Name: Be specific about what you want to extract. Examples:

• “Clause 1 Text”

• “Party Names”

• “Contract Effective Date”

2. Description: Write a detailed instruction for the AI to follow. Examples:

• “Extract the text of Clause 1 from the contract.”

• “Extract the names of all parties involved in the agreement.”

• “Extract the effective date mentioned in the contract header.”

3. Options (Optional): Provide specific choices to guide the model if multiple possibilities exist. Example:

• “Option 1: January 1, 2023”

• “Option 2: February 15, 2023”

Note: After typing an option press "Enter" key to add the option.

Datapoints Example

• Datapoint: Clause 1 Text

Description: “Extract the text of Clause 1 from the contract.”

• Datapoint: Party Names

Description: “Extract the names of all involved parties.”

• Datapoint: Contract Effective Date

Description: “Extract the date when the contract becomes effective.”

• Datapoint: Expiration Date

Description: “Extract the contract expiration date.”

Why it’s important: Each field ensures the AI extracts only the required information, avoiding irrelevant data.

Step 4: Use Multiple Extractors

You can use the results from one extractor in another to handle complex workflows.

For example:

• Extract Clause 1 Text using the first extractor.

• In the second extractor, compare Clause 1 Text with Clause 2 Text from the same document.

Use placeholders to reference results dynamically:

• Press ‘/’ and select the datapoint from the first extractor.

Tip: You can enable Generator to make the Extractor generate new text.

Why it’s important: This enables layered processing, allowing you to handle advanced use cases like clause comparisons or cross-referencing data.

Step 5: Add the Database Axon

The Database Axon stores extracted information.

1. Create a New Database: To create a new database, simply enter a name in the Database Name field and run the Agent. The platform will automatically create the new database and store the processed data there.

2. Use an Existing Database: If you want to use an existing database, click the dropdown menu, select the desired database, and the Agent will add the new data to the selected database.

Why it’s important: Ensures all processed data is structured, organized, and easily accessible for future use.

Step 6: Customize with Variables

Variables let you reuse predefined terms across multiple steps,

1. Press "@" key and select Create New Variable.

2. Name the variable and add a description. The description will replace the variable when used.

Example

• Variable Name: Governing Law

Description: “State of California.”

When inserted into a prompt, this variable dynamically populates with the description.

Why it’s important: Variables save time, maintain consistency, and simplify complex workflows.

Monitor Agents

Track your Agent’s progress on the Monitor Agents page:

• Real-Time Updates: View each step’s status.

• Pause/Resume: Stop workflows for review and continue as needed.

• Retry Failed Tasks: Quickly resolve any issues.

• Access Databases: View structured information in the Databases page for further analysis.

Best Practices

1. Use Lyrics: Save time by describing your workflow in natural language and let Lyrics generate the Agent for you.

2. Reuse Axons: Avoid starting from scratch by reusing existing workflows.

3. Name Axons Clearly: Use descriptive names for easy identification in Monitor Agents.

4. Enable Manual Review: Pause and verify critical steps as needed.

5. Check Monitor Agents Regularly: Ensure workflows are running smoothly and catch errors early.

By following these steps and best practices, you can create efficient workflows that deliver precise results. Remember, clear and thoughtful prompts are the secret to harnessing the full power of 913.ai. Happy building!