Core Concepts
How NEO interacts with your code, data, and ML workflows
Agent Characteristics
NEO operates with four core behaviors that affect how it works in your project:
- Context-Aware – Remembers your code, data, and ongoing tasks during a session.
- Self-Correcting – Detects errors in code, data, or model training and retries automatically.
- Iterative – Refines models using validation results without manual intervention.
- Transparent – Writes logs of decisions, code changes, and reasoning to the VS Code output panel and project folder.
How It Works
1
Send Task to NEO
Type your request in plain English in the VS Code NEO panel:
“Build a classification model to predict customer churn”2
NEO Analyzes Project
Automatically inspects your code and data folder, then plans the steps:
- Checks input datasets and formats
- Chooses model type and features
- Prepares preprocessing and training steps
3
Run Pipeline
NEO executes the pipeline in VS Code:
Data → Validate → Feature Engineering → Train → Evaluate
4
Review Results
Outputs appear in your project folder and VS Code panel:
- Trained model files (Pickle, ONNX)
- Evaluation metrics and plots
- Feature importance reports
Supported Data Types
NEO supports these data sources:
- Tabular: CSV, Excel, Parquet
- Images: JPG, PNG
- Text: TXT, JSON
- SQL Databases: PostgreSQL, MySQL
- Cloud Storage: S3, GCS, Azure Blob
- Audio: WAV, MP3, FLAC
ML Capabilities
NEO can handle these ML tasks:
- Tabular ML: classification, regression, clustering
- Computer Vision: image classification, object detection
- NLP: text classification, NER, summarization
- LLM Fine-tuning: LoRA, instruction tuning
- GenAI & Agentic AI: RAG systems, autonomous agents