Over 250,000 Infosys employees are trained in GenAI: CTO Tarafdar

Tarafdar spoke how the company has built three million lines of codes using the technology and what is the purpose of its AI-first strategy

MOHAMMED RAFEE TARAFDAR, chief technology officer (CTO) of Infosys, is driving the information technology services firm’s artificial intelligence (AI) strategy. Tarafdar, in a video interview with Shivani Shinde, spoke how the company has built three million lines of codes using the technology and what is the purpose of its AI-first strategy. Edited excerpts:

How is Infosys’ AI-first strategy being implemented and how do you quantify the impact?

AI-first essentially means, can we have AI embedded into every aspect of work and it also creates value. We focused on three areas: How do we amplify human potential; second, how to unlock value and finally; and how do we transform our operating model and talent so that they are future-ready.

To give an instance on amplifying human potential, for developers, especially those who are into coding and testing, we launched a number of code assistants. We have 14,000 developers who are using this for a lot of internal work. We are also using these code assistants in some of our clients' environments, with prior permission.

Similarly, learning for us is big. We have integrated GenAI (Generative artificial intelligence) into our learning platform and rolled out to all 320,000 employees. This means if I want to learn differently in an interactive manner, I can do that right within our learning platform. These are some examples where we are using these to amplify the potential.

What do some of these changes mean operationally?

We realised that we cannot continue to operate in the same way that we were doing earlier. Earlier, we had a digital operating model – now we have evolved it to bring elements of AI into it.

What this means is that if I'm doing process reimagining, how do I do it in an AI-first manner? If I'm doing automation, how do I now bring the AI agents into it? We also recognise that unless all of our employees understand the power of technology…so we are doing talent transformation also. More than 250,000 of our employees are trained in GenAI. We are creating an expert group called AI Builders and Masters. We have at present 6,000 Builders. The idea is that our employees use AI in their day-to-day work and also to improve their productivity.

Quantify the impact of AI being integrated in the firm

The benefits will vary. If you look at coding and testing the benefit is productivity and that varies between 10-35 per cent. Second is the cycle time of processes, so in case of knowledge management and about creating content earlier what took weeks can be done within a week today.

How does this help in improving the quality and efficiency of services? This matters in operations. In some of our campuses we are using AI to forecast energy demand. We are also using it to predict equipment failure.

Infosys recently said that it has built 3 million lines of code. What does it mean in terms of competency?

First it means we have been using this for a long time. For one of our code assistants we have been a zero-day customer. Second, what it also means is that our developers are much more comfortable in using a lot of these AI assistants in order to deliver their day-to-day work.

The third part is that with 3 million lines of code we also understand which activity can use this more effectively.

How many GenAI deals is Infosys working on?

We are working on more than 200 projects in the GenAI space. These are in about eight broad areas. These include sales and marketing, customer service, business operations, IT operations, software engineering, risk and compliance, contact center, and employee productivity and learning. Interestingly we are also finding some use cases which are very business specific to their domain. We are finding that while these are the eight areas that are there, there are also core business specific areas where we are starting to see an uptick.

As firms begin to scale up AI deployment, how important is change management?

The change management aspect of GenAI is real. When we were rolling out the code assistance to reach this 3 million line of code target, we found just giving employees a code assistant doesn't mean they will use it effectively.

What we did was, deploy some mentors into the different projects teams. These were mentors and coaches who would guide them and make it easier for them to understand the value and demonstrate its working. We also created a lot of AI ambassadors and the idea was that these are people who would talk about how they use AI in their day-to-day activities.

We see scale in two ways. One is in terms of the number of users who are using it, which is the mass deployment.

Second is in terms of the areas where it is being deployed. We think the coverage will increase across a number of different areas. So I talked about eight areas that generally we are seeing an uptick. We think that the spectrum will increase and the number of users will also increase.

How do you tackle hallucinations in the models that you create, responsible AI etc?

At the beginning of our AI-first strategy, we were very clear that responsible AI cannot be an afterthought. One of the pillars in our AI-first journey was to be responsible by design. As part of this we identified 12 areas. These include areas like explainability, transparency, traceability, auditability, then the impact on IP contracts and others. We recently also got certified. But it's a very evolving area. We have set up a dedicated responsible AI practice team within Infosys and this team just focuses on all the things that have to be done and we have integrated into our risk management.

Infosys also is focused on creating LLMs ((large language models, the core software of a new artificial intelligence system)

When we say LLMs we are not building a pre-trained model. We are using an Open Source model and fine-tuning it with our own organisational data. We are creating a set of models for our own internal usage and second of course is to create for customers.

We are using it for our own products and platforms and also for areas like SAP etc. We are also using it for testing scenarios. We have created a million test cases. We are also using it for modernisation of legacy systems. For Customers we are helping them implement these fine-tune models.

Will Infosys computing power go up?

Specialised models do not require large compute (computing power). We can create specialised and fine-tuned models in a fraction of cost and compute. Of course we do need compute (computing power) but not like the large models.

Second, we are focusing on smaller scale models of say 2 billion parameters models or 1 billion parameter modules.

We are now starting to see a million parameter models that are very effective.

Having said that as a company, about four or five years back, we realized the power of AI and we invested in building our own AI cloud. We partnered with Nvidia and we set up a cluster of the GPUs to do it. Now, we have to scale the capacity. We also work with the hyperscaler infrastructure.

In the augmentation that we have done recently, we have taken the latest processors from Nvidia and we have augmented as well. We are looking at multiple other players on the infra side that are coming up with.