Hello, what can we help you find?

No, the basic package is alert+consultation only.

Gen AI re-elaborates the underlying context. In the case of Daitomic Chat the context is limited to regulatory documents, so the chat is able to rework this content depending on the prompt received.

We do not use our clients’ data to train our algorithms, which are trained with internally produced data. However, the Daitomic Chat gives the possibility to the user to express a positive (thumb up) or negative (thumb down) feedback which we take into account when working on a new development of the Gen AI model.

(Internal note: this approach could change overtime)

The key to freeup answers of hallucinations is the machine-readable format that we have available inside our platform. Basically, as soon as you type a question, our GenAI chat is searching for the most relevant regulatory context that is needed to answer the question. The user can double check the content that was used to answer the question, to be always sure about the correctness of the information behind the answer. Considering only information that are coming from regulations, the GenAI chat can’t have hallucinations cause the answers is based on precise information

Yes, we use GPT-4o APIs.

We measure the precision with the best practice of academic research. You can find some of our academic work in peer reviewed journals in the public Google Scholar database at the name of our founders. The precision of our AI models are below the 97% as whole measurements. (Internal note: Dev team can better argue on that, also consider that we are iterating on an internal measurement system).

We continuously improve our AI algorithm by means of annotations made by our compliance and legal experts. From a technical point of view, we do not apply any interpretations, our algorithms are trained on the basis of objective parameters such as the hierarchy of regulatory sources and texts’ morphology: regulators uses specific forms when formulating an obligation, a sanction or a definition – our AI models understand this modus scribendi, not only keywords (es. “must” or “shall”). We have a pipeline of continuous delivery of AI models which from one side performs automatic training activities in the light of new data (internal note: this training is automatic because it is unsupervised because once the model has been trained the first time, each time a new document is processed the data increase; nevertheless we have a human-in-the-loop process represented by a team analyzing data and recognizes when there is a sufficient sample to improve the performance by doing a new release) and from the other avoid regressions (internal note: a regression is represented by performance shortfall of between 2 releases, es. today we successfully classify an obligation but after a new release that obligation is not anymore classified correctly, even in the case in which the %level of the performance is unchanged) by performing continuous testing and enriching the testing dataset with a new gold standard for each release.

Our proprietary AI models exploits the way the regulators write down regulation content. The characteristics of legal documents is that the way something is written is characterizing also its semantic meaning: our AI models are trained to recognize the special word sequences that represents an obligation or a sanction, without making any interpretation of the content.

Our platform also supports soft law. We also monitor consultations and Q&A documents. The possibility to integrate answers to consultations also depends on the way in which the specific Authority manages this kind of publication so we should evaluate the specific case.

Yes, because the pricing is not related to the implementation but to the prioritization of the integration of the relevant issuing bodies. That’s because we have a roadmap for the expansion of our database so the possibility to intervene on our roadmap to change the priorities comes at a cost.

Yes. We implement a classic multi-tenancy approach, on the basis of which we create separate tenants for each user thus isolating the private documents. However our mission is to create the “regulatory google” so we want to give access to all customers to our entire database.

(Internal note: We will of course re-evaluate this approach when we enter into a new industry, to make sure our users can easily exploit our search engine without “noise” coming from regulations not relevant for them). 

Our proprietary technology is language independent and some features, such as the Daitomic Chat (our Gen AI technology), make it possible to produce output independently from the input language (es. I could use the Daitomic Chat to converse with a document in English by asking questions in Italian and receiving the answe in French), the front-end instead as of today is English or Italian only but we have already in pipeline the translation in the relevant language. However, a specification is needed for the AI models, because there are some features that are not generalizable, for instance external references.

That’s why when we enter for the first time in a new country we need to train our algorithms to ensure the same level of performance: that’s an activity we can easily do and which we did when we got our first foreign customer (from Luxembourg). In any case we can easily retrieve english versions if available.

(Internal note: in the moment in which we will be active in different countries with different languages, each language will have a separated pricing)

We are open to evaluate integrations to other systems. Our approach is not to give our APIs but to work together with our clients in order to make sure the users can benefit the most from our platform by implementing it in their daily routines and operations. However, on the basis of our experience, even in case of opening our APIs, every integration is an ad hoc project where feasibility analysis is needed, particularly to understand how to integrate every piece of information. To be noted that our partnership (for instance Deloitte-Archer) can ease these activities.

Daitomic provide access to original regulatory sources. We do not provide legal advice. However, we can help you excel using our platform guiding you through the configuration process of special legal inventories or other information that are relevant to your business.

We host our server on AWS, in Ireland (Europe).

We have a proprietary AI model which exploits open source technologies: our platform is based on a patent already issued in Italy (in 2023) and under evaluation in Europe and US. More specifically, the patent has as object the whole methodology used to transform documents in a machine-readable-format.

Few minutes.

We can update the documents for you as soon as you publish a new version of them. There are two main ways: i) generate an automatic upload information flow and ii) periodically send update processed by our operations team and uploaded into the platform. For the current customers, we have agreed on a maintenance cost that they pay annual to update all the internal policies continuously. The same is for taxonomies.

Yes of course we can. However, to have it available in the impact analysis, you should have this taxonomy to be connected to the internal policies. The more they are connected in a granular way, the best precision of impact analysis you will receive back.

It depends on the item to be exported, for instance a legal inventory containing a high number of documents is the item which takes more time to be generated and it takes less than 10 minutes.