HeatWave GenAI

You can use the in-database LLMs to help generate or summarize content based on your unstructured documents. Users can ask questions in natural language via applications, and the LLM will process the request and deliver the content.

A user is asking a question in natural language “Can you generate a summary of this solution brief?”. The large language model (LLM) processes this input and generates the summary as output.

You can combine the power of generative AI with other built-in HeatWave capabilities, such as machine learning, to help reduce costs and obtain more accurate results faster. In this example, a manufacturing company does so for predictive maintenance. Engineers can use Oracle HeatWave AutoML to help automatically produce a report of anomalous production logs and HeatWave GenAI helps to rapidly determine the root cause of the issue by simply asking a question in natural language, instead of manually analyzing the logs.

A user asks via HeatWave Chat “What is the main problem in this collection of logs? Provide a two-sentence summary.”. First, HeatWave AutoML produces a filtered list of anomalous logs based on all the production logs that it continuously ingests. Then HeatWave Vector Store provides additional context to the LLM based on the logs knowledge base. The LLM takes that augmented prompt, produces a report, and provides the user with a detailed answer explaining the issue in natural language.

Chatbots can use RAG to, for example, help answer employees’ questions about internal company policies. Internal documents detailing policies are stored as embeddings in HeatWave Vector Store. For a given user query, the vector store helps to identify the most similar documents by performing a similarity search against the stored embeddings. These documents are used to augment the prompt given to the LLM so that it provides an accurate answer.

A user asks via HeatWave Chat “Which laptops can I order and what is the process?”. HeatWave processes the question by accessing internal policy documents housed in HeatWave Vector Store. It then provides an augmented prompt to the LLM that can generate the response “Here is the list of approved vendors and the steps to follow to order.”

Developers can build applications leveraging the combined power of built-in machine learning, generative AI, and vector store to deliver personalized recommendations. In this example, the application uses the HeatWave AutoML recommender system to recommend restaurants based on the user’s preferences or what the user previously ordered. With HeatWave Vector Store, the application can additionally search through restaurants’ menus in PDF format to suggest specific dishes, providing greater value to customers.

A user asks via HeatWave Chat “What vegan dishes do you suggest for me today?”. First, the HeatWave AutoML recommender system suggests a list of restaurants based on what the user previously ordered. Then, HeatWave Vector Store provides an augmented prompt to the LLM based on the restaurants’ menus that it houses. The LLM can then generates a personalized recommendation of dishes in natural language.

Similarity search focuses on finding related content based on semantics. Similarity search goes beyond simple keyword searches by considering the underlying meaning instead of only searching the applied tags. In this example, a lawyer wants to quickly identify a potentially problematic clause in contracts.

A lawyer asks via HeatWave Chat “In which contracts do we have this sentence?”. HeatWave Vector Store performs a similarity search and provides the answer “This sentence appears in the following 6 contracts.”