OCI AI Services
Service-per-modality map
OCI AI Services are pretrained, API-driven models - you call them without training a neural network yourself. Map the input to the service:
- Text → OCI Language: sentiment, key phrases, entities, language detection, translation, text classification (including custom models), and PII detection/masking.
- Images → OCI Vision: image classification, object detection, OCR/text detection; pretrained or custom-trained models.
- Audio → OCI Speech: batch and real-time transcription, custom vocabulary, text-to-speech.
- Documents → OCI Document Understanding: OCR plus key-value extraction, table extraction and document classification for invoices, receipts, passports and similar forms.
- Numeric time series → OCI Anomaly Detection (multivariate, trained on normal data) or OCI Forecasting (future values with confidence bounds).
Custom vs pretrained
Many services offer both. Pretrained works immediately on general content; custom training needs labeled examples in the expected format and improves domain accuracy, for example a custom text classifier or a custom document model for your invoice layout.
Generative AI on OCI
OCI Generative AI provides a playground, hosted and fine-tunable models, dedicated AI clusters for isolated throughput, and agents that retrieve over your documents.
Traps
Questions test service boundaries: reading fields off a scanned form is Document Understanding, not plain Vision. Anomaly detection requires a training dataset of normal behaviour - it is not zero-configuration. Forecasting requires historical time series.
Rule of thumb
Ask two questions: what is the data modality, and do I have labels? Modality picks the service; labels decide pretrained vs custom. Confirm regional availability and limits before designing.