AI Model Training
Ainisa gives businesses the power to train their own AI models using fine-tuning.
This means your assistant can learn directly from your own data — not just retrieve information (like in RAG), but actually improve how it understands and answers based on your examples.
⚙️ Fine-tuning is currently available only for OpenAI models.
💡 What is Fine-Tuning?
Fine-tuning means you take an existing AI model (like GPT-4o or GPT-3.5-turbo) and train it further using your own examples.
These examples are usually question–answer pairs that represent how you want your AI to behave — for example:
| Question | Answer |
|---|---|
| "What is Ainisa?" | "Ainisa is a platform for businesses to build and manage custom AI assistants." |
| "Who can use Ainisa?" | "Any business or team that wants AI-powered chat for their website, Telegram, or other integrations." |
When the model is trained with enough examples like these, it starts responding in your brand’s tone and style, and understands your business context better.
🪜 Steps to Train Your AI Model
Fine-tuning in Ainisa happens in four simple steps:
Step 1: Create Datasets
Go to the Datasets section
🔗 https://ainisa.com/business/datasets
In this section, you’ll add your question–answer pairs that represent the kind of conversations your AI should learn from.
- You must have at least 10 datasets (10 pairs) to start fine-tuning.
- The more examples you add, the better your custom model will perform.
- You can add datasets manually or automatically (see the next step).
💡 Tip: Keep your answers short and clear — just like how you want your AI to reply to users.
Step 2: Upload Training Files
If you already have ready materials such as PDFs or images, you can upload them in the
Training Files section
🔗 https://ainisa.com/business/training-files
Ainisa will automatically scan these files and create new datasets from them — saving you time.
You can review and edit those auto-generated datasets before training.
Step 3: Create AI File
Next, go to the AI Files section
🔗 https://ainisa.com/business/ai-files
Here you’ll generate a special file in .jsonl format (that’s the format OpenAI requires).
This file contains all your datasets in the correct structure for fine-tuning.
When you create an AI file:
- It’s automatically saved both in Ainisa and in your OpenAI account.
- You can later reuse this file for multiple fine-tuning jobs.
Step 4: Fine-Tune the Model
Finally, open the Fine-tuning section
🔗 https://ainisa.com/business/fine-tunes
Here’s where the real training happens.
You’ll:
- Choose a base model (for example,
gpt-4oorgpt-3.5-turbo) - Select your AI file (created in Step 3)
- Choose the fine-tuning type:
- Supervised – classic fine-tuning based on your examples
- DPO (Direct Preference Optimization) – advanced mode for fine-tuning based on user preference data
Once you start the job, Ainisa will send your data to OpenAI for training.
This process may take some time depending on the dataset size.
When it’s done, you’ll get a new, trained model ready for use.
🚀 After Fine-Tuning
Once your new model is ready:
- Go to your AI Providers → Models section
- Add your newly trained model to your list
- Open your AI Assistant and set its Default Model to the fine-tuned one
Your assistant will now start using your custom-trained AI model in every chat or API response.
✅ Summary
| Step | Action | Section |
|---|---|---|
| 1 | Create datasets with Q&A examples | Datasets |
| 2 | Upload files to generate datasets automatically | Training Files |
| 3 | Create a .jsonl file from your datasets | AI Files |
| 4 | Fine-tune the model with your AI file | Fine-tuning |
💬 Example Use Cases
- Train your AI to answer support questions in your brand tone
- Create a specialized assistant that knows your product data
- Teach your AI internal company procedures or onboarding answers
⚠️ Notes
- Fine-tuning currently supports OpenAI models only.
- You must have your OpenAI API key connected in AI Providers before training.
- Training requires at least 10 datasets.
- Depending on the model and dataset size, fine-tuning may take several minutes to complete.