Skip to content

Core Concepts

Understand the fundamental building blocks of IOValence.

1. What is an AI Agent?

An AI Agent is a software system that: - Receives input (questions, data, images) - Processes using trained models - Provides output (answers, predictions, actions) - Learns from examples

Types of Agents

Classification Agent

Categorizes inputs into predefined classes.

Example: Email classifier (spam/not spam) 

Input: Email text
Processing: Analyzes language patterns
Output: "Spam" or "Not Spam"

Regression Agent

Predicts continuous numerical values.

Example: House price predictor 

Input: House features (size, location, year)
Processing: Learns price patterns
Output: Predicted price ($350,000)

Detection Agent

Identifies objects or patterns in data.

Example: Image object detection 

Input: Photo
Processing: Scans for objects
Output: List of detected objects + positions

Generation Agent

Creates new content based on patterns.

Example: Text generator 

Input: Topic prompt
Processing: Generates coherent text
Output: Generated content

2. Training Data

The foundation of every agent. Quality data = better agent.

What is Training Data?

Examples your agent learns from: - Features: Input information - Labels: Correct answers - Quantity: Typically 100s to 1000s of examples

Example Training Dataset

feature1,feature2,label
5.1,3.5,setosa
7.0,3.2,versicolor
6.3,3.3,virginica

Data Quality Rules

✅ Clean - Remove errors and duplicates
✅ Complete - Fill missing values
✅ Balanced - Equal examples per category
✅ Relevant - Directly related to task
✅ Sufficient - Enough examples to learn

Training vs Test Data

Aspect Training Data Test Data
Purpose Agent learns Measure performance
Usage During training After training
Typical Split 80% 20%
Should overlap? NO - never!

3. Models & Training

What Happens During Training?

1. Agent receives training data
2. Model learns patterns
3. Adjusts internal parameters
4. Repeats until performance plateaus

Key Training Concepts

Epoch: One complete pass through training data - More epochs = more learning (but can overfit) - Typical: 5-100 epochs

Batch Size: How many examples processed at once - Larger batch = faster, less memory efficient - Smaller batch = slower, more accurate - Typical: 32, 64, 128

Learning Rate: How quickly model learns - Higher = faster but can miss optimal solution - Lower = slower but more stable - Typical: 0.001 - 0.1

Overfitting vs Underfitting

Underfitting:           Good Fit:            Overfitting:
Model too simple        Just right           Model memorizes data

❌ Low training acc     ✅ High training acc  ✅ Very high training acc
❌ Low test acc         ✅ Good test acc      ❌ Low test acc

4. Model Evaluation

How do we know if our agent is good?

Key Metrics

Accuracy: Percentage of correct predictions 

Accuracy = (Correct Predictions) / (Total Predictions)
Range: 0-100%
Goal: High accuracy

Precision: Of predicted positives, how many correct? 

Precision = True Positives / (True Positives + False Positives)
Goal: Reduce false alarms

Recall: Of actual positives, how many found? 

Recall = True Positives / (True Positives + False Negatives)
Goal: Don't miss important cases

F1-Score: Balance between precision and recall 

F1 = 2 × (Precision × Recall) / (Precision + Recall)
Range: 0-1 (higher is better)

Confusion Matrix

              Predicted
            Positive  Negative
Actual  Pos   TP        FN
        Neg   FP        TN
  • TP (True Positive): Correct positive prediction
  • FP (False Positive): Incorrect positive prediction
  • TN (True Negative): Correct negative prediction
  • FN (False Negative): Incorrect negative prediction

5. Deployment

Making your agent production-ready.

Deployment Options

Option Best For Complexity
Local Development, testing Simple
Server Small-medium load Medium
Cloud High scale, high availability Advanced
Edge Real-time, offline Advanced

Deployment Steps

  1. Validate - Ensure metrics meet standards
  2. Package - Prepare agent for distribution
  3. Deploy - Upload to target environment
  4. Monitor - Track performance in production
  5. Update - Retrain with new data periodically

6. The Full Workflow

┌─────────────────────────────────────────┐
│     1. Create Agent Configuration       │
└──────────────────┬──────────────────────┘
                   ↓
┌─────────────────────────────────────────┐
│     2. Prepare & Upload Training Data   │
└──────────────────┬──────────────────────┘
                   ↓
┌─────────────────────────────────────────┐
│        3. Train Model                   │
│        (IOValence handles this)         │
└──────────────────┬──────────────────────┘
                   ↓
┌─────────────────────────────────────────┐
│     4. Evaluate Performance             │
│     (Check accuracy, metrics)           │
└──────────────────┬──────────────────────┘
                   ↓
              Performance OK?
              /         \
            YES          NO
            /              \
           ↓                ↓
       DEPLOY          ADJUST SETTINGS
       (USE)           GO BACK TO STEP 2

Next Steps

Questions? Check FAQ →