Automated Legal Research Through Customized Private AI Chatbot

Challenge Story

Secure and Private Chatbot for the Client

Our client, a Canadian law firm specializing in business law services, sought to streamline document access and reduce manual effort in document retrieval, given the extensive volume of documents and files they manage. They aimed to leverage AI technology and transition from a disjointed repository system to a streamlined data search and retrieval solution.

Their specific requirements included an AI chatbot capable of summarizing documents, drafting emails, and performing various AI-assistant functions. The primary focus was on rapid data summarization and maintaining conversational flow while adhering to responsible AI development standards to ensure accuracy and traceability of information.

Emphasizing security, the client needed all chatbot interactions to remain confidential and secure, with no data stored in public databases or employed for future model refinement. Additionally, they aimed to seamlessly integrate the chatbot into their Microsoft Teams channel.

The client received a chatbot for efficient data retrieval that met their efficiency and reliability requirements. Within the span of four weeks, Adastra provided the client with a chatbot, powered by Power Platform and OpenAI, that serves as a search tool for their IT and HR data.

Solution Story

Private, Secure, and Informed Chatbot Built Using Azure OpenAI

Adastra prioritized client security by leveraging Azure’s OpenAI services to set up a personalized private instance. The client received a custom chatbot created using Microsoft’s Bot Framework Composer to directly address their use cases.

To ensure that Azure OpenAI API responses maintained appropriate context collected from previous interactions, inquiries, and inputs, the client collaborated with Adastra to implement context retention capabilities.

Deployed within the client’s existing Azure Web service, the solution was accessible through a shared URL, seamlessly launching the chatbot within Microsoft Teams via an integrated channel. Future plans include further integration within Teams.

During implementation, Adastra streamlined the proof of concept (PoC) by segmenting document content for manageability. Initial accuracy issues were resolved by increasing the overlap between information segments, adjusting bot configuration to prevent inaccurate responses, and implementing indicators for unanswerable queries.

The scalable solution, currently functioning as a chatbot PoC within Microsoft Teams, encompasses approximately 60 documents, with expansion plans in development.

Benefits Story

Secure Communication, Improved Customer Services & Future AI Possibilities

• With Adastra and Microsoft’s support, the client experienced:
• 90% reduction in data retrieval effort, boosting productivity and saving costs.
• Uninterrupted availability.
• New opportunities to implement OpenAI.
• Maintenance of chatbot context capabilities.
• Scalability for faster response times and workflow optimization.
• Enhanced engagements.
• Secure interaction with private endpoints.
• Streamlined access to organized electronic data archives to research past cases.

The solution utilizes Azure OpenAI for embeddings, Chatbot 3.5 for large language models (LLM’s), Azure AI Search for indexing, and the Microsoft Bot Framework Composer for the Chatbot interface. Additionally, the client also received app service resources and an Azure bot within Azure’s ecosystem.

This solution enables the client to explore extensive electronic data archives and conduct meaningful research on past cases during the discovery phase. This capability allows them to identify relevant documents and information essential for legal proceedings.

By the end of the project, Adastra helped the client unlock the full potential of AI and AI chatbots. The client intends to expand the solution in the future to incorporate legal information from an existing repository.