How we automated project requirements management using AI and ML

In this article, we want to share the story of how we developed a product using modern AI, ML, and NLP technologies and applied this product to optimize our own processes. The main problem is compliance with requirements: there are always a lot of them! Especially in the case of designing and operating complex aggregated composite systems produced by various subcontractors. In the course of work, it is necessary to take into account many both minor and significant conditions of the requirements, each of which can affect the final result. Therefore, great importance is attached to tracking, updating, and verifying all requirements at each stage of the project. This is especially important in projects where exact compliance with the technical specifications is required. Having investigated the problem, we found that the task of compliance and tracking of technical requirements is also acute for technical personnel (project offices, engineers), lawyers, economists, and company management.

All participants need to:

• reduce the risks of non-fulfillment of contractual obligations;

• link the requirements of one contract with the requirements of another (trace requirements), to improve the manageability of requirements and reduce the risks of "unaccounted" or forgotten requirements in the contract;

• reduce the time for analyzing project contracts, financial events;

• account for the connections of financial events, requirements, and obligations;

• reduce the time for high-quality digitization and import of technical documents into requirements management systems;

• consider financial risks and non-fulfillment of obligations.

We are integrators in business and interact with our customers, and accordingly operate with their set of project documentation, which includes, at a minimum, regulatory legal documentation and EPC contract and ITT, we also develop our technical specifications for systems based on these documents, and use the composite subsystems of our subcontractors based on our technical specifications, which also have their private technical specifications and technical conditions. We will decipher everything a little below.

General hierarchy of technical project documents:

• Regulatory - legal documentation (GOSTs, Laws)

• EPC contract (E-engineering, P-procurement, C-construction) and ITT (initial technical requirements)

• Technical specifications (TS)

• Private technical specifications (PTS)

• Technical conditions (TC)

And if we follow the terms of GOST 59194-2020 "Requirements Management Basic Provisions" more precisely, there is a need to trace all the connections of requirements between objects and all their components.

Can you imagine how labor-intensive the volume for engineering and technical analysis is? :)

Our customers from the project office had an "urgent" task at that time - to iteratively perform full traceability for all technical requirements for a large number of project objects starting from high-level documents and in descending order of priority:

• ITT \ TZ

• TZ \ CHTZ

• CHTZ \ TU

Our team came up with the idea of creating an intelligent assistant that could automate the process of requirements management in terms of establishing traceability (connections) to control all the requirements presented to the system, reducing the workload on employees and minimizing the risk of human error. Such products exist in the foreign market, but there are not many analogues in the Russian market. This is due to the complexity of developing such products. But our team had enough expertise to solve this problem.

We had a task for automation: to match the project technical requirements to the system with the technical requirements to the subsystem and establish a connection.

We proposed a product that assists the expert in analyzing the requirements of these documents. The product first in automatic, and then in manual (finishing\controlling) mode with the possibility of using a "Smart Assistant" helps the user - the project expert to establish all the connections of the requirements contained in these documents.

We used a stack of AI NLP technologies and ML, starting from "Classical" approaches to text comparison based on Jaccard similarity using the N-gram method, enhancing comparison mechanisms by applying text vectorization algorithms, semantic search, as well as machine learning for deeper adaptation to project data.

The result of our work was a product that helped our customers achieve traceability of all requirements 6 times faster than the planned time!.

In addition, the time for checking and updating requirements is significantly reduced in case of project changes or versioning, which ultimately speeds up the entire process of developing and implementing large projects.

We received feedback, conducted problem interviews, presented the product, and identified additional necessary functionality for the intelligent tracing tool. This includes automatic parameterizable text atomization and document entity classification to identify requirements, financial events, and obligations, and their specific interrelationships. Active work is currently underway to refine this functionality, including the implementation of AI methods in these processes.

Thus, the use of advanced AI, ML, and NLP technologies and our internal expertise have not only facilitated our work but also significantly improved the quality and efficiency of project activities in other company departments. We are confident that our experience can be useful to other organizations facing similar challenges in requirements management, and we are now conducting pilot implementations of the system.

If you are interested or have any questions, please write to [email protected].