AI Training
Last updated
Last updated
At RAIA, we believe great AI isn’t just smart—it’s informed. That’s why our platform is built to train each AI Agent with the right combination of general intelligence and business-specific knowledge, resulting in conversations that feel accurate, helpful, and personalized.
Here's how RAIA makes that happen.
Every RAIA Agent starts with a powerful foundation: OpenAI’s industry-leading large language models. These models offer exceptional reasoning, language understanding, and general knowledge—out of the box.
But here’s the magic: RAIA lets you layer on your internal knowledge and shape your AI's behavior with tools that make it uniquely yours.
To ensure RAIA Agents deliver high-quality, context-aware responses, we use a 4-layer architecture to control what they know and how they respond:
Powered by OpenAI This is the base layer. It allows your Agent to answer general questions, use logic, and write in natural language. But on its own, it doesn’t know your company, your customers, or your specific processes.
Your documents, embedded into your Agent
This is where RAIA becomes your AI. You can upload documents like:
SOPs, product manuals, policies, FAQs
PDFs, DOCs, TXT files
Training decks, meeting notes—even website pages via scraping
RAIA automatically converts this content into semantic embeddings and stores it in a vector store. This allows the AI to retrieve relevant answers based on your exact wording, tone, and expertise.
How to upload training content:
Go to the Agent → Train tab
Upload files (max 20 docs, up to 500MB total)
Optionally tag or title each document for organization
RAIA syncs and indexes them instantly
🧠 Want dynamic updates? You can enable integrations with cloud storage like Google Drive for automatic syncing.
Controls how the Agent thinks and responds
This is the "personality and mission" layer of your AI Agent. You define how it behaves, what its boundaries are, and how it solves problems.
Example instructional prompts:
“You are a helpful assistant trained to support customer success reps at a B2B SaaS company.”
“Always escalate billing issues to a human agent.”
“Use a casual, supportive tone—like you're a smart friend helping out.”
Fine-tune each conversation with user-specific context
This is where true personalization happens. The Context Prompt lets you inject dynamic, session-based info such as:
The user’s name, plan level, or purchase history
Support ticket metadata
CRM or internal system data
This prompt is passed at runtime via:
API (in the body of a POST request to the conversation endpoint)
Live Chat SDK (embedded context when launching the widget)
💡 This means your AI Agent can greet users by name, tailor answers to their account, or change tone depending on user type—all in real time. You can define this during agent creation in Launch Pad or pass it dynamically via the API or Live Chat SDK for specific use cases.
Training doesn't stop at setup. RAIA’s Copilot experience lets you test your AI Agent in real conversations and improve it continuously with human feedback.
Rate Responses with 👍 or 👎
Leave Comments explaining what was right or wrong
Suggest Better Answers directly in the UI
Flag Threads for future training use
Admins can also add specific threads to training memory, improving the Agent’s responses over time—no code needed.
By combining these four layers—General AI Model, Vector Store, Instructional Prompts, and Context Prompts—RAIA ensures your AI Agent is:
✅ Smart (thanks to OpenAI) ✅ Aligned with your company’s voice and rules ✅ Informed by your internal documentation ✅ Personalized to each user interaction
It’s everything you’d expect from a well-trained employee—only faster, more scalable, and always available.
Head to Launch Pad to create your agent, upload documents, define your prompts, and watch your AI workforce grow smarter with every interaction.
