Making Data Conversational with AI
Why This Matters
- The customer is a large radio broadcaster in Italy with 5 million users, representing 10% of the Italian population.
- The customer has over 50 years of legacy data, including structured tables, equipment manuals, regulatory information, and coverage maps.
- The customer had previously tried to address the data challenges with a relational database solution, but it had significant limitations and was no longer in use.
The Data Challenge
- The customer's data consisted of:
- Structured tables for equipment maintenance and issues
- Unstructured equipment manuals
- Regulatory information
- Coverage maps of the radio station's broadcast area
- The previous relational database solution could not effectively handle the diverse data types and resulted in significant information loss.
The Solution
- Knowledge Base: The team built a knowledge base using Amazon DocumentDB, a vector database, to store and manage the customer's diverse data.
- Retrieval Augmented Generation: The team implemented a retrieval augmented generation architecture to enable natural language queries, rather than SQL, to access the data.
- Map Analysis:
- The team used a large language model (LLM) with vision capabilities, such as SONET 3 or Amazon NOVA, to extract metadata from the coverage maps and generate a semantic description of the maps.
- This allowed the maps to be searchable using natural language queries.
- Deployment and Impact:
- The map-based solution was deployed first, demonstrating the transformative potential of generative AI capabilities.
- The customer is now in the process of adding the other data sources, such as equipment manuals and regulatory information, to the system.
- The switch from SQL to natural language queries has made the application more accessible to non-technical users.
- The solution has significantly reduced the time to retrieve information, from days to minutes, which is a game-changer for the customer.
Key Takeaways
- Generative AI can unlock new possibilities for working with diverse data sources, including unstructured data and complex assets like maps.
- Combining retrieval augmented generation with vector databases can enable natural language-based access to data, improving usability and accessibility.
- Unlocking the value of legacy data through innovative solutions can lead to significant business impacts, such as faster time-to-value and new capabilities.
If you're interested in learning more or discussing the use case further, you can reach out to the presenter, Francesco Ciitti, via email or LinkedIn.