Return to landing page

Functional overview legalneer.ai

What is legalneer.ai?

legalneer.ai is an AI-based platform that uses LLMs (Large Language Models) to efficiently support lawyers and students in their daily legal work. LLMs are AI models that have been trained on large amounts of text data and can understand and process human language. Among other things, they enable the automatic analysis, summarisation and generation of texts. We provide tools for research, summarising and automated document review. Use of legalneer.ai by non-legal professionals is expressly not intended.

legalneer.ai accesses data uploaded to the MyData database, ensuring complete transparency of the sources used so that users can verify the relevance of the information provided. The focus is on the targeted use of your own data, enabling legalneer.ai to optimise the quality and efficiency of legal work processes. Using your own documents ensures that the answers are based on well-founded, user-specific sources. This avoids the problem that large language models for the legal field are often insufficient due to a lack of technical depth.

With legalneer.ai, users optimise their legal work and use their expertise more efficiently. Users can ask targeted questions of the specialist database, create drafts of legal documents and create additional value by optionally sharing data sets.

A decisive advantage is the high level of data control, which enables users to fully control and manage their own database. With the generated answers, legalneer.ai offers an initial assessment of legal issues, based on relevant sources. In practice, this can save a considerable amount of time.

Intensive testing with legalneer.ai has shown that specialised LLMs offer significant added value for legal applications. Precise prompt formulation and optimally coordinated parameter settings enable realistic results to be achieved that can come close to the level of a sample solution. The user interface of legalneer.ai is also convincing due to its appealing and clear design. The application is self-explanatory and makes it easy to get started, so that even users without extensive technical knowledge can work with it quickly and productively. legalneer.ai is based on an interdisciplinary research project and is still in the development phase. Although the AI-generated results offer many advantages, they may contain errors and must therefore always be checked by a subject matter expert for accuracy and completeness. One of the challenges is the lack of reproducibility of the generated texts, as they are never created with identical content, which makes it difficult to validate the results. Added to this is the possibility of hallucinations and bias, which can result in inaccurate or distorted answers. For this reason, it is essential that the results are carefully checked by subject matter experts to avoid errors or misinterpretations in the argumentation and in the result.

Our AI-powered research project meets high standards in terms of data security, transparency and versatility. We ensure the protection of your data in accordance with European and German regulations, provide full disclosure of our information sources and flexible, individual solutions. Users are explicitly responsible for the uploaded data, in particular with regard to legal admissibility, such as in relation to copyrights and other relevant legal provisions. Unless users explicitly decide otherwise, the uploaded data is reserved exclusively for the user himself. However, there is an option for users to share their own data with other users, while retaining full control over what data they share and under what conditions In accordance with Art. 50 of the AI Regulation, legalneer.ai also ensures that users are informed in a clear and comprehensible manner that they are interacting with an AI system. In addition, the AI points out that the generated content may contain potentially erroneous or biased results.

To use legalneer.ai, simply go to https://legalneer.ai and create an account. legalneer.ai is currently in the development phase and is continuously being improved. The platform offers a wide range of applications for various legal tasks – from answering specific legal questions and preparing expert opinions to researching laws and legal texts and checking contracts using AI.

As a non-commercial project, legalneer.ai's sole aim is to support professional users and companies in optimising their work processes. We only charge the cost price for the use of these services, without making any commercial profit. In addition to our standard solutions, we are also open to customised solutions for companies based on research and development collaborations.

If you have any comments, questions or require further information, please do not hesitate to contact us at info@legalneer.ai.


Applications


Setting the model parameters

General settings

In the general settings, you can select the desired LLM model. Each of these model variants offers different performance characteristics, which may vary in speed, accuracy and resource requirements.

  1. Gpt-4o: Standard model with balanced performance and accuracy, suitable for general applications.
  2. Gpt-4o-mini: Resource-efficient and faster, ideal for less complex tasks with slightly reduced accuracy.
  3. Gpt-4-turbo: predecessor of GPT-4o

Advanced settings

  1. LLM temperature: A slider adjusts the creativity of the responses: A high value (e.g. 1) leads to more creative output, while a low value (e.g. 0.2) makes the output more predictable and deterministic.
  2. Max. number of output tokens: In addition, the maximum number of output tokens can be set to determine the maximum length of the answers.
  3. Number of sources for RAG (k): The k-value setting can be used to determine how many relevant sources are included in the pre-selection for generating the answer. The value can be set between 1 and 30. The default value is 20 sources.
  4. Search type for sources: In the first step, sources relevant to the RAG must be searched for. Two search types can be selected for this, Similarity and MMR.
  • Similarity: This method measures the similarity between the retrieved sources and the question or context.
  • MMR: This method combines similarity with diversity to optimise the selection of sources. The user's choice of search type influences how the model evaluates and selects sources, with similarity focusing on precise matches and MMR simultaneously focusing on diversity and redundancy avoidance to provide more comprehensive and diverse results.
  1. Retriever Type: The sources found in this way are either no longer ordered (Base) or reordered/reweighted using different approaches (Reordering and Rerank).

  • Base: This is the standard retrieval method. The model accesses the base database of documents without reordering the sources.

  • Reordering: This option reorders the results. After the information has been retrieved, the order of the sources is adjusted to prioritise results that may be more relevant.

  • Rerank: With Rerank, sources and the associated question are sent to an external language model, which re-evaluates them and can thus significantly increase the quality and relevance of the results. In addition, Rerank offers the option of reducing the number of sources actually used. This is done using the setting "Number of sources RAG after rerank step" (top-k), where the least relevant sources are removed. The value for top-k must always be lower than the original k value.