Flame is flagship AI platform developed by Atalgo Engineering to enable enterprises to utilize computational intelligence to solve complex business problems. The Flame’s philosophy is to build a computational model of an enterprise as an operating environment for technical and business process to be AI augmented. This is a massive undertaking and our our approach is iterative. In first iteration, we want to build a platform where organizations can build a Knowledge Base of the various processes. We are calling this Intelligent Knowledge Engineering (iKE). In second phase, we will enable our users to develop and deploy intelligent applications to do various tasks more autonomously, efficiently and with minimal supervision. The platform will learn as it will be used across various functions within an organization. A high level roadmap of Flame AI platform is depicted in below diagram –

Flame knowledge engineering capability (iKE) is in beta testing and we are aiming to formally launch it by end of March. We have done extensive testing on various specialist domains and are getting wonderful results. The platform does reasoning on the knowledge artefacts provided within Flame Admin and provides accurate answer. We have the capability to do calculations and generate response based on the context and previous questions. Flame platform provides cloud based ability for end users to interact with Knowledge Artefacts securely and can support multiple users within organization using the platform for various queries. The Project -> Process -> Knowledge Base model of Flame platform mimics how information is structured and accessed within an organizations with various access levels. Flame has the capability to scrape information and builds a model of various toolsets such as Jira/Confluence, Jenkins, CRM etc. We are adding more platforms in our base model and Flame also provides a template to add any custom tool within the organization. We natively support Windows which can help end users automate any processes on their machine. Overall Flame provides a comprehensive capability to manage complex knowledge artefacts and allows end users to reason on them and find answer to complex problems.

Flame Admin – Flame Admin is a central place for knowledge engineering. It allows an organization to build Knowledge Bases, create models of various tools and platforms, create projects and allow users access to different projects depending upon the organizations needs. It has powerful UI driven workflow for complete knowledge engineering. There is an audit capability of various activities being undertaken by different users.

Flame Client – Flame Client is the interface through which an end user can interact with organization’s knowledge artefact, query in natural language and develop workflows to automate many tasks in the background. This is a powerful utility which allows users to utilize Flame’s generative AI capabilities to find insights and automate various processes. Users can initiate multiple sessions and parallelize their tasks to actionable insights and achieve significant efficiency in their tasks.

Sample Test – As part of our beta testing phase, we are testing Flame capabilities extensively over complex and specialized domains such as International Water Laws, Drug Dosage Calculations and Quantum Mechanics. We feed Flame various knowledge artefacts and sources, develop a benchmark set of questions to test Flame’s reasoning ability, context awareness and certain other parameters and test Flame against those benchmarks. We are getting encouraging and exciting results and Flame is doing substantially better on calculation and complex reasoning tasks than ChatGPT. I am describing below the drug usage use case and the result which impressed me great deal. We pointed Flame knowledge base to a publicly available https://www.georgebrown.ca/sites/default/files/2020-05/Drug%20Dosage%20Calculations.pdf for drug dosage calculation and asked the questions given in the sheet. Flame answered all the questions correctly which involves mathematical calculations in the background. I am pasting the results of the interactions as below –

Question – The doctor orders 90 milligrams of liquid cough syrup. The liquid cough syrup has a label that reads 120 milligrams (mg for short) in 5 milliliters (or mL for short). How much cough syrup should the nurse give to the patient?

Flame Response – Correct Answer

Question – Ampicillin 500 mg capsules are supplied. MD orders 1.5 g. How many capsules should be given to the patient?

Flame Response – Correct answer (this requires multi step calculation hence getting answer right without additional prompt was impressive)

Question – The doctor orders “ibuprofen 600 mg PO BID”. You have 300 mg tablets of ibuprofen on hand. How many tablets should be given to the patient at one time?

Flame Response – Correct answer

Question – The order says, “erythromycin suspension 600 mg PO q6h.” The supply on hand is erythromycin 400 mg per 5 mL. How many milliliters of medication should be given to the patient?

Flame Response – Correct answer

Question – The physician ordered “penicillin V potassium 400 000 units PO QID”. You have penicillin V potassium 200 000 units per 5 mL. How many milliliters should be given to the patient?

Flame Response – Correct answer

This is just a sample of multiple results we have got from our testing in this domain and others. What impressed me most is Flame’s ability to utilize generative AI capabilities to formulate correct sentence while applying calculations and reasoning to find the accurate answer.

We are just scratching the surface as far as future possibilities of our AI platform is concerned. In parallel, we are working on a development environment where Flame engineers can build complex intelligent workflows and autonomous intelligent programs to support IT and business tasks.