Improvement & Roadmap

If new scenarios or document templates appear, does the system learn automatically, or is retraining/tuning required?

Manual tuning or recalibration of the base model is typically not required.

The system continuously improves through a feedback loop mechanism rather than retraining the foundational model. When users correct metadata, adjust classifications, or add missing information, those inputs feed into a “self-healing” prompt framework that refines extraction logic and decision rules over time.

This approach allows the system to auto-adapt to new variations in templates and evolving document patterns while maintaining the stability of the core model..

Is prompting a continuous process? Are prompts continually created, and how does that relate to self-learning?

We release baseline prompts. For new doc types, we create prompts and train on your docs. Prompts self-heal via feedback loops and improve accuracy over time.

Can Automate automatically link documents with the EDL and update it accordingly?

This capability is not yet fully available but is currently in development.

EDL management is more complex than individual document processing because it requires contextual understanding across multiple documents and structured cross-validation to ensure accuracy and completeness.

Roadmap Context:

TransPerfect’s development roadmap follows a logical progression:

  1. Individual Document Processing : Classifying documents, identifying issues, and extracting metadata from a single document.

  2. Cross-Document Intelligence : Evaluating documents in relation to one another (e.g., completeness, sequencing, timeliness, and expected pairings).

  3. EDL Management : Understanding how a newly uploaded document impacts the broader TMF, including fulfilling placeholders, reconciling expected document lists, and dynamically updating the EDL.

The EDL management functionality is currently under active development as part of this phased enhancement strategy.

Can the model automatically link documents with the EDL, fulfill placeholders, and update the EDL?

Not yet. This capability is on the product roadmap and currently under development.

EDL automation is more complex than single-document processing because it requires cross-document context and validation across the TMF structure to ensure accuracy and compliance.

Roadmap Progression: (1) single doc processing, (2) documents in relation to one another (order/completeness/timeliness), (3) EDL management.