Priority access to download (fast track) is provided to professional emails of gov domains, NGOs, and large companies.
All key features are free to use. ( Frioul version 1.2)
- Creation of classification, extraction, or combined IDP models.
- Unlimited annotations (document and extraction labels).
- Model training and inference (unlimited trainings, inference limited to 5,000 pages).
- Image correction (orientation, skew, noise...).
- scoring : readability scores for images, confidence scores for classification and data extraction.
- User arbitration interface: you can correct the output according to a confidence limit.
- Performance analysis interface by labels.
- Retrieval of extracted data (.csv).
- splitter : automatic splitting of pieces on trained documents.
- splitterAI : agnostic zero-shot piece splitting (alpha).
- quickLearnAI : similarity classification engine.
- fewShotAI : few annotations, thanks to generative AI applied to annotated documents, to multiply performance.
- quality : module to validate the quality of extracted data, according to criteria for each data type.
- API access.
Technical prerequisites
The AI solution is designed to work on a server architecture with GPU.
Recommended configuration: 32-core CPU, 128GB RAM, 1TB SSD storage + 2TB HDD, RTX A6000 GPU (48 GB).
However, feature evaluation can be done, with degraded performance, with a standard server configuration.
A setup documentation is provided with the solution.
Recommended configuration: 32-core CPU, 128GB RAM, 1TB SSD storage + 2TB HDD, RTX A6000 GPU (48 GB).
However, feature evaluation can be done, with degraded performance, with a standard server configuration.
A setup documentation is provided with the solution.
The following features (which are either in pre-production or require configuration assistance) are not deployed in the free version:
- rights: advanced user role and rights management module for the platform.
- reTrain: intelligent model retraining planning.
- autoImproveAI: the model automatically self-feeds to improve performance.
- arbitrageAI: the arbitration interface includes advanced features such as model retro-improvement, the possibility to arbitrate files (session of several documents), workflow replay, suggestion of new labels on unrecognized documents, correction on historical documents.
- flows: you build in-app advanced data processing workflows.
- compute: you build calculation rules for data validation.
- lookup: universal connector for querying third-party services (API, SQL).
- truthOrDareAI: document authenticity validation module (image forensics).
- fraudAI: fraud detection models (documentary inconsistency, intra-file inconsistencies, inter-file suspicions) based on business rules.
Contact-us if you are considering an experiment (POC) or if you wish to meet an approved integration service provider.
