Train a Model

🧠 Philosophy

While large language models (LLMs) are excellent for reasoning, logic, and generalization, there are many use cases where structured data and classic supervised learning still outperform — especially when accuracy, explainability, or tabular features are critical.

INTELLITHING supports both. The Train a Model section lets you quickly train and deploy traditional machine learning (TML) models from structured data (CSV), with zero code — and then use them as intelligent building blocks alongside LLMs in your workflow.

Our approach is simple:

• Upload or select your dataset
• Clean and preprocess your data
• Choose a task (regression or classification)
• Let AutoML build your model
• Use the trained model as a drop-in block anywhere else

Whether you're a data scientist prototyping, or a domain expert automating workflows, model training is as easy as clicking Next.

🔑 Key Concepts

Concept	Description
Dataset	A CSV file with features and a target column
Preprocessing	Data cleaning, encoding, normalization, etc. done before training
AutoML	A built-in system that chooses the best model architecture and hyperparams
Regression	Predict continuous values (e.g. price, temperature)
Classification	Predict categories or labels (e.g. fraud/no fraud)
Model Block	A trained model becomes a block you can use in the visual workflow editor

📘 Key Definitions

Term	Meaning
Target Column	The label you're trying to predict
Features	The input columns used to predict the target
Train-Test Split	Automatically applied split to validate model quality
Cleaned Dataset	Intermediate version of your dataset after preprocessing
Block Output	The prediction(s) made by your trained model during runtime

🧩 Workflow at a Glance

One-line Summary

CSV → Preprocess → AutoML → Model Block → Workflow Ready

Visual Walkthrough

1. Upload or select your CSV dataset
2. Preprocess: remove outliers, fill missing values, scale, encode, etc.
3. Click AutoML or pick Regression / Classification
4. Choose your target and features
5. Click Next — training starts
6. The trained model appears as a block in the editor

⚙️ How Model Training Fits into INTELLITHING

Model training is not a separate ML platform — it’s part of your workflow.

• When you train a model, it becomes a reusable block
• The block takes structured input, predicts an output, and passes it to the next step
• You can mix TML blocks with LLMs, tools, indexing, and more
• It’s a seamless blend of logic, prediction, and orchestration

🚀 How to Train a Model

Step 1: Upload or Select a Dataset

• Go to Train a Model
• Click Upload a dataset or choose from existing CSV files
• You’ll see the preview and last modified time

Step 2: Preprocess Your Data

Once a dataset is selected:

You’ll see options like:

Action	Purpose
Remove Outliers	Drop extreme values that may skew training
Drop Columns	Exclude irrelevant or high-cardinality features
Fill/Remove Missing	Handle missing data via fill or deletion
Remove Duplicates	Clean duplicate rows
Binning	Bucketize numeric variables
Polynomial Features	Generate feature interactions automatically
Normalize / Scale	Standardize data for numerical stability
Encode	One-hot or label encode categorical columns

📝 You can apply multiple actions before proceeding to training.

Step 3: Choose a Task

After preprocessing, click:

• AutoML (recommended): INTELLITHING selects the best fit

• Or choose manually:

• Regression: For predicting continuous numbers

• Classification: For predicting categories

Step 4: Select Target & Features

• Pick your target column (what you want to predict)
• Select input features (what the model should use)
• Click Next to begin training

⏳ The training runs server-side — no setup required.

Step 5: Your Model Becomes a Block

Once trained, your model is automatically saved as a block:

• Appears in the Block Editor
• Can be dragged into any workflow

• Has standard inputs/outputs:

• Inputs: Features as dictionary

• Outputs: Predicted values or class

You can describe the block (inputs/outputs) like any other and combine it with other tools or LLM blocks.

🛠️ Model Block in Workflows

After training:

• Go to Block Editor
• Drag your model block into any workflow

• Use it like any other block:

• Input = Dictionary of features

• Output = Prediction (number or label)

✨ This is how you blend TML precision with LLM flexibility — for example:

• Use a model to score or rank candidates
• Use an LLM to explain the prediction
• Use a rule to route based on confidence

🔐 Permissions

Action	Required Role
Upload Dataset	All authenticated users
Train Model	All authenticated users
Use Model Block	All users with access to the project
Modify Preprocess	Creator or editor of the dataset

💡 Best Practices

Tip	Why It Helps
Clean before training	Garbage in = garbage out
Use AutoML first	Fastest path to strong baseline
Be intentional with target	Choose the business variable you're optimizing for
Use feature names clearly	Makes model input/output more readable
Test models in workflows	See how they interact with LLMs or triggers

💬 Summary

• Train a Model lets you go from CSV to deployed model block in minutes
• Use preprocessing to clean and transform your data
• Choose from AutoML, Regression, or Classification
• The trained model becomes a block, reusable across workflows
• You can now combine TML with LLMs, rules, and tools in one automation system