How to Build an AI Agent
🚧 IMPORTANT: This Tutorial is currently in development
🚨 Have additional questions? Please open a ticket or ask on our Discord.
1. Create your Base System Prompt
![](https://docs.stammer.ai/~gitbook/image?url=https%3A%2F%2F1359281993-files.gitbook.io%2F%7E%2Ffiles%2Fv0%2Fb%2Fgitbook-x-prod.appspot.com%2Fo%2Fspaces%252FkLIQnOFYHkQtdWxUzzFE%252Fuploads%252FZMysFK666k2lW0WIcQ7V%252Fimage.png%3Falt%3Dmedia%26token%3D8669e237-301c-476c-8211-f7ef5fdfa9c4&width=768&dpr=4&quality=100&sign=f8168258&sv=1)
The base system prompt is the foundation of an AI agent, defining its behavior and capabilities. To create an effective prompt:
IMPORTANT: Start with the default prompt provided by Stammer (see below)
Make small, incremental changes to fine-tune the agent's performance.
Use clear, specific language to convey instructions and guidelines.
Provide examples to illustrate expectations and context.
Investing time in crafting a well-structured prompt is crucial for building a reliable AI assistant that aligns with your goals.
The Stammer Default AI Agent Prompt:
Select Your Model (GPT-4o Recommended)
![](https://docs.stammer.ai/~gitbook/image?url=https%3A%2F%2F1359281993-files.gitbook.io%2F%7E%2Ffiles%2Fv0%2Fb%2Fgitbook-x-prod.appspot.com%2Fo%2Fspaces%252FkLIQnOFYHkQtdWxUzzFE%252Fuploads%252FnssUaB9I8waOQwOW6zTw%252Fimage.png%3Falt%3Dmedia%26token%3Dd8713c34-b799-42af-832e-5e6e9ba46b8d&width=768&dpr=4&quality=100&sign=6708935c&sv=1)
Choosing the right model for your AI use case is crucial:
GPT-4o: Recommended for most use cases, especially complex tasks like lead generation and scheduling. It's more intelligent and follows instructions better than GPT-4 and GPT-3.5.
GPT-3.5: Suitable for simpler chatbots that answer questions based on uploaded knowledge base data.
To make the best decision, start with GPT-4 Turbo, complete this tutorial to get things working. Then, temporarily switch to GPT-3.5 to see if it meets your needs.
Write your prompt in English and then ask the Agent to translate
For optimal AI model performance, it's best to provide instructions in English. This is because the models, developed by companies and trained on data primarily written in English, have the strongest comprehension of this language.
To cater to non-English speaking end-users, we highly recommend:
Write your base system prompt in English.
Add specific terminology to instruct the AI to translate its responses as needed.
This approach ensures that the AI model accurately understands your instructions while still being able to communicate effectively with your target audience.
Using Markdown can be helpful to better structure your prompt to provide additional context but this is OPTIONAL
Using markdown formatting can be incredibly helpful when creating complex prompts. Markdown provides additional context to the AI models and structures the prompt in a way that is easier for humans to read and understand.
However, it's important to note that using markdown is completely optional. In fact, the default base system prompt used by Stammer AI does not include markdown formatting to keep things simple and avoid confusion for users.
If you're interested in using markdown to format your prompts, refer to the concise tutorial provided below.
2. Connect Your Data Sources
Upload data to the AI agent's knowledge base.
📚Knowledge Base Explained3. Get the AI Agent Answering Questions accurately FIRST
Connect your Data Sources
Remember, you must get clean data into the bot
Get the A.I. Agent Answering Questions Accurately FIRST
Use Debug Mode if anything is not answering properly
If the bot is not answering correctly then you need to:
Get better data into the bot
Add the question to your data with the appropriate answer
Setup Lead Collection OR Appointment Scheduling
Lead Collection
Appointment Sched
Understand limitations
Consider alternative link based approach
That is all folks!
Last updated