Flame Intelligent Knowledge Engineering

Flame provides enterprise grade intelligent knowledge engineering capabilities across the organization which can act as foundations for the reasoning of AI systems. Atalgo engineering team has been doing extensive testing of its knowledge engineering capabilities across various complex domains such as Quantum Mechanics, Complexity Science, Healthcare, Finance etc and results are encouraging. This has also helps us find challenges around how Flame manages the corner cases and deals with the hallucinations/inaccuracies generated from raw LLM output. We also have the capabilities for users to provide feedback and Flame will re-generate the output with added information until the user is satisfied. Also, dynamic knowledge management provides capability to keep enhancing the knowledge base while end user is using the platform.

We have recently added exciting capabilities in our knowledge engineering module, such as support for multiple documents and document types. Flame’s robust architecture support multi-step workflows. The key parts of Flame knowledge engineering component are –

-????????? Decision Node for where to search Knowledge Vector space

-????????? Web Relevancy Checker Node to check relevancy between user query and retrieved documents

-????????? Hallucination grader Node for grading generated output as with or without hallucination

-????????? Answer Grader Node to grade answers whether Flame answers the user question or not

Flame architecture ensures that learning received by the platform from the information provided is adequate and error free; and can be computationally processed by AI workflows. We are in the process of quantifying and formalizing the learnings and intelligence of the programs/workflows developed on Flame platform.

Here, we are providing the details of one of the testing carried out to showcase Flame’s ability to process and understand complex mathematical formulae in a pdf document. We did this test on Quantum Mechanics domain where we created few knowledge artefacts with one having Quantum Mechanics formulae attached as pdf (link - here)

We asked a few questions which tested Flame’s ability to process the formula in the pdf and generate a summarized response in plain text. Some of the examples are given below –

The next two questions test Flame’s ability to summarize mathematical formulae in natural language just by ingesting the pdfs.

We are continuing to test Flame’s ability to act as a secure and scalable knowledge engineering platform for AI systems. We are continuing to test its ability to summarize the contents in the context of questions and aims of the project an avoid inaccuracies and hallucinations. Also, as this platform is under active development, we are continuously making progress into the new features; especially around creating intelligent workflows and providing a full-fledged development environment to Flame users to be able to develop AI applications. If you want to gain access to the platform or contribute in further development an testing, please fill out the form at https://www.atalgo.com/flame