Sure, the AgileDD document classification is very robust and extremely adaptable because it is based on human supervision.

With AgileDD you can define the categories that fit the needs of the end users.

That means that legacy classification can be used to accelerate the training process.

Secondly, the AgileDD’s classification process takes into account that a document like an End of Well report may contain several types of information.

Therefore, it is possible to associate several tags to a document and also ask for several tags per document at the time of the inference.

In addition to the classification of documents, the AgileDD platform can also classify the document contents at the scale of a page.

The figures and tabular data can also be classified by our platform.

Yes, our tabular data extraction is extremely reliable because the human is in the middle of the loop to guide it.

The capture of tabular data implies the detection of the table, the classification of the table according to their contents and the segmentation of the table to detect the cells and associate them with a column header and a line header.

The human is in the loop to guide the table classification, improve the segmentation either manually by downloading table table into Excel or trying several automatics segmentations available in the platform.

Absolutely! The AgileDD solution provides an LLM-based chatbot for you to query your documents in natural language.

Thanks to an advanced search tool and the RAG technology (Retrieval Augmented Generation), it is possible to query one or several documents with some very satisfactory replies.

AgileDD makes possible the use of commercial and very large LLM such as ChatGPT or Gemini only available on external APIs or opensource LLM as Mistral, Llama, Falcon or Bloom which can be installed locally and eventually locally fine-tuned.

Using the AgileDD chatbot you will be fascinated with the possibility to discuss some subsurface topics with all the geologists and engineers who have been working on you asset before!

Thanks to its advanced user interface, it is very easy to train supervized models to detect target information from the well log header.

Our platform will allow:

• -Classify the logs per category
• -Detecting the well name and well metadata like the country, the county, the well head coordinates, the well total depths (driller and logger), the spud date …
• -Detect the logging service name and associated metadata like the logged intervals, the logging date, the run numbers, the unit number, the log mnemonics, the logging company, etc.

This capture of key information can be done using either a LLM or a highly focused and accurate supervized model fine-tuned locally by the data experts.

AgileDD is agile! This platform is made-up of a set of APIs used for the UI and all the AI tasks.

Using the APIs you can build workflows to extract the information from documents using existing models or to access pre-captured information and to output into structured formats such as CSV, JSON formats or SQL input records.

Maturing a model may be a continuous and iterative process which is important to monitor. Versioning the models and knowing their main characteristics is also very important.

Therefore, we have developed a model governance tool to access the characteristics of the models such as their training data, the performance (precision, recall, F1-score) or the way they converge in the case of a computer vision model.

All this information is important for the project manager to select the right model version, the one which has the best performance for it needs.

Definitively yes.

The geology of the well is frequently reported as a log with text description and/or graphical intervals. AgileDD can be trained to capture the graphical pattern describing the intervals, the symbols of shows and to detect the geological descriptions.

The descriptions can be analyzed to recognize the lithology, the petrophysical characteristics and the structural information.

As with any capture of information done by the AgileDD platform, the information is perfectly sourced by its location in the document and can be exported along the well depth axis.

Once injected into our platform, you can directly start exploring your data set and be guided for this type of “needle in the haystack” data capture.

At first you can use our advanced search which not only uses the full-text index of your documents but also integrates some filtering options on the classification, the text or graphical detections you may have done earlier.

You can also continue the search using the chatbot and get more detailed and sourced occurrences of the mineralization you are looking for.

Thanks to the RAG (Reinforced Augmented Generation), advanced prompt for In Context Learning, your information capture strategy will be based on some geological reasoning and not only keyword filtering.

If needed, this can be completed by fine-tuning a generative AI model to specialize it run very accurately this type of search with a very high level of accuracy and exhaustivity.