Improve customer satisfaction with Generative AI at a leading South African University
Legal | Customer Service| Generative AI
Case Study Details
Leverage Generative AI to service and empower your customers to resolve their own queries, freeing up your employees for more value added tasks.
Challenge
Our client needed a legal advice solution to assist health research scientists with the correct interpretation of the applicable country specific laws for the handling of sensitive health research data and its movement between countries across the continent.
Solution
Lightblue proposed employing a Generative AI, Retrieval Augmented Generation (RAG) solution architecture to best meet the customers’ requirements. The RAG approach is accepted as one of the best practices at mitigating hallucinations by the Large Language Model.
The RAG med involves making use of a vector database to store embeddings which enable semantic search of relevant content snippets to be passed to the Large Language Model with the end users’ question for a natural language response to be generated.
Approach
- Business Requirements: Detailed business requirements were gathered and documented from stakeholders. This step required sign-off before proceeding to ensure alignment with business objectives and expectations.
- Solution Architecture: The solution architecture was designed, defining the technical blueprint and system integration requirements. This included selecting the appropriate technologies and frameworks for the project.
- Conversational Platform – Instantiation, Integration, Configuration, and Extension (Customisation): This task involved setting up and customising the conversational platform. Necessary components were integrated, settings configured, and functionalities extended to meet specific project needs.
- Large Language Models – Pipeline Engineering: Focus was placed on pipeline engineering for large language models, ensuring data flowed smoothly through various stages of processing.
- Integration Testing: Integration testing was conducted to verify that all system components worked together seamlessly. Any issues arising from the interaction between different modules were identified and resolved.
- User Acceptance Testing: User acceptance testing was carried out to ensure the system met the end-users’ requirements and expectations. Feedback was gathered and used to make necessary adjustments before deployment.
- Knowledge Transfer: Knowledge transfer was critical for ensuring the client’s team could effectively use and maintain the new system. Comprehensive training and documentation were provided to facilitate this process.
- Deployment: Deployment involved launching the system into the live environment. This step included final checks and ensuring all components were correctly configured for operational use.
Results
The implementation of our Generative AI legal advice solution yielded remarkable results:
- Enhanced Compliance: Health scientists could ensure better compliance with diverse and complex data protection regulations across different African countries. The AI solution could provide tailored legal advice, helping scientists navigate and adhere to local and international data usage laws, reducing the risk of legal issues and penalties.
- Improved Data Governance: By providing consistent and accurate legal advice, the AI solution could enhance data governance practices. This would lead to more secure and ethical handling of sensitive health data, protecting patient privacy and maintaining trust.
- Efficiency and Productivity: Automated legal advice significantly reduced the time and resources spent on legal consultations. Health scientists could quickly get the information they need, allowing them to focus more on their core research activities and accelerating the pace of scientific discoveries.
- Standardisation of Legal Practices: The solution could help standardise legal practices related to data usage across the continent. This would facilitate collaboration and data sharing between different research institutions and countries, fostering a more integrated and efficient research environment.
- Cost Savings: Reducing the need for extensive legal consultations and minimizing the risk of non-compliance penalties results in significant cost savings for research institutions and health organisations.
Lessons Learnt
- Iterative Development: Adopting an iterative development approach, such as Agile, allows for flexibility and adaptability. Regular feedback cycles and incremental releases help address issues early and ensure the solution evolves in line with user needs.
- Customisation and Localisation: Customising the AI solution to accommodate local legal frameworks and languages across different African countries is critical. This requires a deep understanding of regional legal nuances and cultural contexts.
- Access to External Knowledge: By accessing a large repository of external documents during the generation process, RAG ensures the information provided is based on verified sources. This reduces the chances of the model generating incorrect or fabricated details.
Conclusion
This project underscores the transformative power of Generative AI to support health scientists in managing their legal responsibilities more effectively, ultimately contributing to more ethical, compliant, and efficient health research across Africa.