For search and generative AI projects, the Search & Generative AI orchestration framework is composed of a collection of technology-independent components. These can help improve search & generative AI development for repeatable use cases, accelerate project timelines, and reduce overall project costs.
The framework provides a REST service created in Python (Search API) which is well documented with Swagger. It has a pipeline architecture optimized for search and generative AI applications which makes it easy to customize functionality to handle complex generative AI requests through the addition or customization of individual stages.
The Search & Gen-AI user interface is the accompanying UI for the framework, made in React and having common functionality found in many search UIs like dynamic fields, facets, filtering, pagination, sorting, highlighting, type ahead, "did you mean", search analytics. etc. It also contains many useful user interface components for generative AI, such as content comparison, semantic search, question-answer, and a “chatty” (e.g. chat-bot like) interactive dialog interface.
Available Generative AI components
Saga Query
It allows integration with Saga to enable the usage of On-prem language models as well as all of the out of the box Saga NLP algorithms to process natural language content.
It allows to calculate embedding vector from the query for semantic search, it can be used with Saga or directly with models like: Open AI and Sentence Transformer GTR.
Prompts (documentation to be public in the next week)
Use Azure OpenAI chat-completion models (like gpt-35-turbo or ChatGPT) to answering user requests and queries with summaries, content classification, data extraction, etc. You design and configure the prompts and include whatever metadata or content from the user query to generate answers.
It is the executor of the semantic search in the search engine to fetch relevant data that could provide context for the prompt creation.
Use Cases
It will flatten the user interface
- The user will just type what they want.
- This will replace multitudes of complex nested menus, tabs, dialogs and navigators.
- New features can be implemented much more quickly and easily as independent plug-ins and add-ons.
Computers will be expected to do much more complex, end-to-end tasks
- Users will want to be able to make complex, multi-step requests with natural language.
- The computer will be expected to guide the user through those steps, prompting them along the way
The search box now becomes the single-most critical user interface component
- It is, today, the only place where people enter text to get what they want.
- It will become the first & primary method by which users interact with your application.
Expectations on language understanding will grow dramatically
- Users will expect it to have human-like interactive capabilities.
- This includes answering complex questions and helping you to achieve complex tasks.
Search API although it is a very complete template, it is NOT a final project. Its main objective is to speed up the initial process of creating a project. The code base is maintained, enhanced and delivered out of the box, but always requires initial configuration.
Search API is the new successor of the Enterprise Search UI, an API framework built with Python 3.9+
The key features are:
- Python 3.9+: Coded in Python, to reach a broader audience of programmers.
- FastAPI: Web framework with a high performance, on par with NodeJS
Truly Engine Agnostic: Add new Non-SQL engines without disrupting the rest of the code
UI Independent: Worked without an UI, and generic enough to adapt one to it
Pipeline Framework: Execute complex process of functional modules, editable on runtime
True Http Endpoints: Enable E2E management of HTTP request for custom process
- PyQPL: New PyQPL integrated for complex query generation (1.0.6)
- Authentication: Local, LDAP, SAML, Delegated already implemented, more as demanded
- Validation: Communication with the API with JWT for UI and API Keys, for S2S
Built-In Documentation: API via Swagger and configuration data via Pydantic
Space contributors
{"mode":"list","scope":"descendants","limit":"5","showLastTime":"true","order":"update","contextEntityId":828178642}
Overview
Content Tools