ELA (Easy To Learn AI)

Confusion matrix
The metrics interface
The Report
User interaction (adding a file, specifying a target, selecting/deselecting features, and choosing the problem type)
Webpage interface
download model

ELA — Easy to Learn AI

From Raw CSV to a Leak-Free, Audited, Deployable ML Model — Locally.

Machine learning is powerful, but building a reliable workflow requires knowledge of preprocessing, validation, evaluation, and deployment. For beginners, these steps are often where mistakes happen.

ELA turns raw tabular data into a complete, reliable ML workflow while keeping the important decisions visible.

Upload a CSV, select the target and features, and ELA handles:

Automatic feature detection
Imputation, scaling, and encoding
Leak-free preprocessing with Scikit-Learn Pipelines
Classification and regression
5-fold cross-validation with GridSearchCV
Model evaluation and error analysis
Local AI-powered auditing
PDF experiment reports
Deployable ".pkl" model pipelines

Leak-Free by Design

ELA splits the data before preprocessing.

Imputation, scaling, and encoding are fitted only on the appropriate training data through "Pipeline" and "ColumnTransformer".

This prevents test-set information from leaking into training and makes the automated workflow more trustworthy.

AI Auditing Without the Cloud

After training, ELA uses Ollama + Qwen3:8B locally to audit the experiment.

The audit examines model performance, errors, confusion matrices, class imbalance, potential bias, weaknesses, and edge cases.

The dataset stays on the user's machine.

From Experiment to Deployment

ELA doesn't stop at evaluation.

The complete preprocessing and trained model pipeline can be exported as a ".pkl" file and loaded directly into Python applications, APIs, or other software without retraining.

Built for Understanding

ELA is not designed to hide machine learning behind a single button.

It is built around four principles:

«Automate the repetitive. Expose the important. Prevent common mistakes. Keep data local.»

The complete workflow:

Raw CSV → Preprocessing → Training → Evaluation → AI Audit → Deployable Model

Built With

Python · Flask · Scikit-Learn · Pandas · NumPy · Ollama · Qwen3:8B · Joblib · ReportLab · JavaScript · Bootstrap

What's Next