Flame – AI Augmented Problem Solver

A powerful AI platform will be expected to solve complex problems efficiently and effectively.

Flame as a world changing AI augmented problem solver is expected to work at the level where it can help enterprises and individuals solve complex problems using machine learning and intelligent knowledge engineering. The vision of Flame involves behaving like an interactive problem solver capable of taking inputs from domain experts and capable of generating heuristics related to the problem at hand. As a secure and compliant problem solver, we should be able to control what information Flame can have access to (for example whether it can refer to internet or not) to be able to solve the problem.

Steps involved in solving a complex problem by an intelligent agent are below –

  1. Understand the problem domain

  2. Understand the relevant knowledge available

  3. Assess the gap at high level

  4. Generate heuristics from the available knowledge (FKB – Flame Generated Knowledge Base)

  5. Seek feedback on heuristics (refine heuristics)

  6. Flame engineer develops workflow to solve the problems

  7. Generate further heuristics in the direction of solution (FKB)

  8. Execute updated workflow

  9. Repeat 5 through 8 till solution is found

Also another aspect of Flame in the enterprise context is that Flame engineers can create intelligent applications which can autonomously run however their efficiency and effectiveness will need to be continuous monitored to ensure they are behaving as expected against new data.

We will try to understand this using a simple example –

Let’s assume Flame does not have full understanding of addition operation however is capable of guess work given multiple similar operations. let’s assume we ask flame to compute below operation –


Let’s assume Flame has below two operations in its knowledge base –



As soon as we ask Flame “9+11”, it automatically generates below 5 heuristics (in Flame terminology, we will call this Flame Generated Knowledge (FKB)) –


3+6=8 (Flame generated a wrong knowledge artefact here)





Once FKBs are generated Flame engineer reviews them and provides feedback as below. If Flame engineer decides these FKBs are of not much use, it can provide further prompts and ask Flame to regenerate FKBs. This is a key feature of Flame where platform is learning in real time about the problem domain as well as its approach to solve the problem from a domain expert interactively –

2+5 = 7 [5 stars, very useful]

3+6=8 [0 stars, please discard]

0+0=0 [1 star, correct but of not much use]

5-1=4 [4 stars, useful]

9+10=19 [5 stars, most useful]

[1,2,3]+[3,4,5]=[4,6,8] [3 stars, useful]

Based on the inputs from the expert, Flame generates the solution. If Flame engineer is happy with the solution, it can close the solution loop and “deploy” the application for further use or just end the application after getting the answer. Flame engineer can reuse this learning for any similar tasks.


One important feature of Flame will be re-usability of solutions for other problems. Flame will continuously and iteratively learn from the solutions and will allow other Flame engineers to re-use the solutions in other similar problems. The chain of solutions will provide emergent phenomena which can be further utilized for other purposes.