FAQ
Quick answers to common questions about NEO. Contact support: support@heyneo.so.
Quick Navigation
- Getting Started – Setup guide and first steps
- Data & File Handling – Formats, uploads, and storage
- Platform vs Extension – Choosing your deployment mode
- Task Submission – Writing effective tasks
- Technical & Security – Architecture, privacy, and security
Getting Started
NEO supports multiple ML domains:
- Tabular ML – Classification, regression, clustering, time series
- Computer Vision – Image classification, object detection, OCR
- NLP – Text classification, sentiment analysis, NER, summarization
- Audio & Speech – Speech recognition, audio classification
- LLM Fine-tuning – Instruction tuning, LoRA, domain adaptation
- Anomaly Detection – Outlier detection, fraud detection
What is NEO and how does it work?
NEO is an autonomous ML agent automating the full pipeline from data to deployment.
Workflow:
- Describe your ML task in natural language
- Provide data (upload, URL, or cloud)
- NEO analyzes data and selects models
- Receive production-ready artifacts with documentation
Example Task:
Build a customer churn prediction model using customer_data.csv. Optimize for recall since missing churners is costly.
NEO handles preprocessing, feature engineering, training, evaluation, and artifact generation automatically.
Do I need ML expertise?
No. NEO is designed for all skill levels.
| Beginners | ML Practitioners |
|---|---|
| Use task templates | Specify models/constraints |
| Step-by-step explanations | Access detailed reports |
| Start simple, progress | Customize deployments & evaluation |
| Describe your business goal | Modify generated code in VS Code |
The key is clear goal description, not prior ML knowledge.
How long does a typical project take?
| Task Type | Duration |
|---|---|
| Simple tabular models | 15-30 min |
| Image classification | 30-60 min |
| Large datasets (>1GB) | 1-3 hrs |
| NLP fine-tuning | 2-6 hrs |
| Custom deep learning | 4-12 hrs |
Tip: Start with a small sample, then scale.
Data & File Handling
Supported Formats
| Format | Use Case | Platform | VS Code |
|---|---|---|---|
| CSV | Tabular/time series | ✅ | ✅ |
| Parquet | Large datasets | ✅ | ✅ |
| JSON | Structured/log data | ✅ | ✅ |
| Images | CV tasks | ✅ (50MB) | ✅ |
| Audio | Speech/music | ✅ (50MB) | ✅ |
How do I handle large datasets?
Approach:
- Platform – Use cloud storage (S3/GCS/Azure)
- Convert to Parquet – Faster processing
- Test first – Use 10% sample
File limits:
- Platform upload: 50MB/file
- Platform cloud storage: unlimited
- VS Code: unlimited local files
Does NEO handle missing data?
Yes. Automatic detection and imputation:
| Data Type | Strategy |
|---|---|
| Numerical | Mean, median, predictive |
| Categorical | Mode, “Unknown” |
| Time Series | Forward fill, interpolation |
| Advanced | ML-based imputation |
Platform Mode vs VS Code Extension
| Feature | Platform | VS Code Extension |
|---|---|---|
| Access | Browser | VS Code editor |
| Setup | Quick, no install | Install once |
| Data | Upload ≤50MB or cloud | Local + cloud |
| Artifacts | Downloadable | Generated in workspace |
| Code Editing | View only | Full IDE + Git |
| Best For | Prototyping, collaboration | Customization, local dev, large datasets |
Which mode should I use?
Platform Mode: Quick results, no setup, collaborative testing
VS Code Extension: Edit code, work with large local files, full IDE features, version control
Task Submission
| Good | Poor |
|---|---|
| Predict customer churn using customer_data.csv (50k rows). Optimize for precision-recall balance. | Do some ML with my data |
How do I write an effective task?
Include:
- Goal – What to predict/classify
- Data – Files, size, key columns
- Metrics – How to measure success
- Context – Business relevance
What metrics should I use?
| Task | Metric |
|---|---|
| Regression | RMSE, MAE, R² |
| Classification | Accuracy, F1, AUC-ROC |
| Time Series | MAPE, SMAPE, directional |
| Ranking | NDCG, MAP, precision@k |
Map metrics to business goals:
- Minimize false positives → precision
- Catch all fraud → recall
- Balance speed & accuracy → F1-score
Technical & Security
| Feature | Description |
|---|---|
| Data Encryption | At rest & in transit |
| No Sharing | Never shared with third parties |
| Complete Control | Delete or export anytime |
Is my data secure?
Platform: Cloud encrypted, deletion on request, no sharing
VS Code: Local files never leave machine, secure cloud access via credentials, full control
Can I see generated code?
Yes, NEO provides:
- Preprocessing & modeling code
- Step-by-step notebooks
- Deployment scripts
- Documentation & methodology
What if model performance is low?
Improvement:
- Improve data quality/features
- Adjust metrics/constraints
- Provide domain knowledge
- Request specific approaches (ensembles, deep learning)
Focus on practical business impact, not perfect accuracy.
Need More Help?
- Documentation – Full docs
- Use Cases – Real-world examples
- Contact Support – Direct help