Enginuity

Enginuity is a not-for-profit organisation dedicated to supporting employers, training providers and individuals in the UK Engineering sector. Through their comprehensive resources, they strive to create opportunities for growth and development, helping to develop the skills necessary for a successful future in engineering. Their support and guidance offers a range of beneficial services, including educational programs to aid in the understanding of important industry topics and the provision of career advice to help individuals make the most of their potential. With the aim of creating a more informed, engaged and prosperous engineering community, Enginuity are determined in their commitment to providing the necessary resources to ensure the continued success of the sector.

A well known charity within the Engineering and Manufacturing sector, with a mission to encourage career progression in this innovative industry. Primarily this project focuses on natural language processing, extracting embeddings from job, CV, apprenticeships and educational datasets to learn a common skills language. We have tackled a number of difficult problems in the project and feel this is particularly relevant to the goals UCAS could face in version 2 or 3 of this personalisation project.

Value Added

• We are actively working with Enginuity to develop their Data Science Strategy and Roadmap.
• Guidance on technical possibilities and limitations in developing skill models and career progression across a wide range of data sources.
• Delivered workforce foresighting tools and machine learning based gap analysis for their client, Department of Education.
• Developed clustering models and skill similarity algorithms to identify shared job tasks across occupations using O*NET dataset.
• Built backend recommendation system to predict most suitable Vaccine Manufacture role given current Job title or Study Subject Area.

The Deliverable

We have worked across a number of different projects with Enginuity but sadly are not able disclose details of the implementation.

A well known charity within the Engineering and Manufacturing sector, with a mission to encourage career progression in this innovative industry. Primarily this project focuses on natural language processing, extracting embeddings from job, CV, apprenticeships and educational datasets to learn a common skills language. We have tackled a number of difficult problems in the project and feel this is particularly relevant to the goals UCAS could face in version 2 or 3 of this personalisation project.

Value Added

• We are actively working with Enginuity to develop their Data Science Strategy and Roadmap.
• Guidance on technical possibilities and limitations in developing skill models and career progression across a wide range of data sources.
• Delivered workforce foresighting tools and machine learning based gap analysis for their client, Department of Education.
• Developed clustering models and skill similarity algorithms to identify shared job tasks across occupations using O*NET dataset.
• Built backend recommendation system to predict most suitable Vaccine Manufacture role given current Job title or Study Subject Area.

The Deliverable

We have worked across a number of different projects with Enginuity but sadly are not able disclose details of the implementation.

Please note, this is a redacted visualisation of skill clustering from the O*NET dataset.

####################

How AI helps with creating standards in Manufacturing.

The Purpose: showcase experience of applying AI models (NLP) for creating and updating standards in manufacturing (using O*net data)

The audience: HVMC, Enginuity, etc. The skill specialist organisations

Executive Summary

Our client, a skill specialist organisation in the UK, reached out to us seeking technological support in activities of creating and maintaining engineering and manufacturing standards. The main problem was that these activities rely heavily on experts’ knowledge which takes a long time to build. The process of standard creation itself is a painstaking and tedious job. A lot of qualification experts’ time goes into skimming through multiple text statements while searching for valuable content.

To help our client, we created an AI-based technology that works using the latest advancements in the language processing space. It supports qualification specialists from two perspectives. First, it helps to highlight the redundant statements that are not needed in the industry anymore. Second, it finds gaps in a standard by showing what the industry does and what is currently not present in the examined role or engineering activity.

As with a lot of technological solutions, the tool has its shortcomings, and cannot fully replace experts’ judgment and knowledge. But it allows automating the routine text analysis and gives an opportunity to focus on relevant parts. It preprocesses hundreds of pages, searches for resembling parts, and suggests pre-screened material to the expert.

Introduction:

Consistent and reliable operations and production are critical in the engineering and manufacturing sectors. This is the main reason why a lot of engineering processes are standardised. These standards provide a set of instructions and guidance on how to perform a role or a task. Thus, they enable people who received certification to do the work to their best ability. However, technologies advance rapidly and quite often do it across sectors. For example, the use of liquid nitrogen has been borrowed from medicine by manufacturing.
Manufacturers need standardisation to deploy new tools. Delayed responses prevent them from seizing opportunities that these advances bring or cause various risks. The traditional approach involves a manual review of standard documentation by an expert. It is laborious, time-consuming, knowledge-intense, and repetitive in its nature process.

When the expert is gone, knowledge is gone too.

Enginuity, a skill specialist company in the field of engineering, came to us with a request for improving the process of creating standards. Engineering standards were stored in PDF format files with a length of a single document of up to 40 pages. Each standard has specific content, details, and terms. With this complexity, it takes a lot of practical and theoretical knowledge to create an engineering standard. Expertise has to be built gradually through collaboration and practice. The main challenges that the company faced were the following:

1) Transferring Knowledge of creating standards is a slow process. The description of Engineering activities is full of jargon, professional terms, and context. When new joiners start collaborating, it takes a long time to gain a full understanding. And if an expert leaves the company, knowledge loss occurs.

2) The process of creating or updating a standard is repetitive when it comes to reading existing standards, finding relevant statements, updating them with new terms, and vetting out redundant parts. Even cross-sector experts who have broad knowledge struggle in reaching for relevant content.

3) A lot of standards have shared units that are common among them. For example, the “Gas Engineering” qualification shares a unit of statements about safety with “Plumbing and Domestic Heating”. However, to create a new qualification, a creator has to have extensive knowledge of a broad variety of standards (in this case Gas Engineering and Plumbing) or manually search through hundreds of them.

Semantic Search AI tool

Repetitive tasks make a good use case for an AI application. In this case, its role is to provide intelligent automation. Previously, the lack of advanced linguistic models made it difficult to perform complex textual manipulations. The main shortcoming was a contextual meaning: the machine didn’t understand the semantics, only doing well in keyword matching. What is semantics? Let’s say you are searching for users of a forum, typing in the search bar: “users of forum A”. The semantical meaning is that you are actually looking for users, but the search engine returns various pages of that forum barely related to your request. Semantic answers the question: “What is that you are actually looking for?”

However, recent advancements in language processing, such as GPT-3, brought viability to a variety of linguistic tasks. With these, making intelligent (contextual) matches became possible. For our solution, we used one of the GPT-3 models called “text-ada-001” as a core technology for contextual matching of standard statements.

To provide ease of use, our frontend engineers have built a user interface (UI) which is a shell that qualification experts see on their computer monitors. It allows the matching process to be visible on one screen with two parallel sections of text: one from existing job roles and one from emerging technologies tasks. The UI is connected to a database where the tool stores all the data. The GPT-3 is linked to the system via API (application programming interface). API is a way to make calls (literally calls) to another system (in our case GPT-3). Through these calls, the user sends a request in a form of job tasks and receives back a list of closest matches and a list of missing items. That list of missing items is a void that new technology requires.

Pros and Cons

As with many technological solutions, the tool has its pros and cons

• The main benefit is that it reduces the burden for a person to thoroughly read through lengthy repetitive documents in a search for relevant information. Putting a lot of strain on attention span, not only the process takes a lot of time, but also makes it unavoidable to accidentally skip through some points. The tool ensures that none of the important statements from a document will be missed.
• In addition, the tool eases the constraint of prerequisite knowledge. A creator doesn’t start with a blank page. The tool provides suggestions that were matched according to the given context. Thus it allows a new user to take on the task that previously required extensive expertise for skill gap identification. In fact, new users can learn from the tool as it provides initial content.

While efficiently dealing with routine text processing, the proposed solution has its limitations.

• The machine learning model that is at the core has been trained on generic textual data. While capturing general sense quite well, it still doesn’t put much weight on some acronym words that are critical to the task it is reviewing. For example, knowing particular pipe specs might be a crucial item for a gas engineering qualification. But the AI system won’t necessarily pick it up as it hasn’t “seen” it before: it wasn’t trained on pipe specs.
• Another weakness of the system is that it only provides suggestions from a list of existing statements. Thus, the quality of matches is as good as the existing content. If a creator tries to write a standard for new technology, they will be limited by a predefined scope of statements. Thus experts will need to be involved when designing new standards.

Wrap up

Despite the limitations of the solution, it has important benefits. It takes away information overload and refines content. It removes repetitive jobs and routines: things we people are bad at. And it helps to focus on more important parts of finding patterns and discovering trends. The tasks we are good at.