We are developing Indic LLMs to truly democratise technology: Ramprakash Ramamoorthy, Director AI Research, ManageEngine

AI today, is at the heart of techies. Enterprises are racing to be among the leaders in deploying AI solutions effectively not only to boost business output but also to create an impression of being the frontrunner in latest tech adoption. Ramparkash Ramamoorthy, Director AI Research at ManageEngine shares his vantage points on the evolution of AI in the country and how it can be a game changer for enterprises. 

As you are playing a key role in AI research, what do you think are the challenges that enterprises in India face while adopting AI-driven solutions?

The challenge with enterprises is privacy, especially after the onset of large language models (LLMs) there are limited insights on how your data will be used. However, despite privacy-aware machine learning (ML) techniques are available, enterprises face difficulties. 

This takes us to the next challenge. When the same AI stack was given to three different companies, they had three different outcomes on the return on investment (RoI) on the stack. What is that all about? A typical analogy could be, that Tesla’s AI-powered self-driving cars work fine around 80 percent of the time in the US. However, the same vehicle to India will probably not perform despite the same AI stack. This can be attributed to the digital maturity in India versus the US. Also, the road markings, traffic signals, the driving style of people on Indian roads, and more play a significant role in decision-making by the car’s self-driving mode. The same thing applies to enterprises as well. 

One of the key concerns regarding the lack of digital maturity is the inter-departmental digital divide. Departments do not talk to each other, so the data is in silos. Therefore, the AI model is unable to see the holistic picture and hallucinates. Therefore, companies need to streamline their processes and data and invest in automation. This can lead an organisation on the path to digital maturity. It is a pursuit that leads you to improve your chance to reap better ROI on your AI stack. Enterprises can further invest in low-code or no-code tools that can connect all the departments internally plugging the gaps.

You quoted ‘digital maturity’ as a core reason behind reaping a better RoI from AI deployments. So are we not ready to go big with AI?

We have grown in India and it is the third biggest market for ManageEngine. Further, our cloud business has recorded 70 percent growth year-on-year in India. Soon, it will be the second biggest market for us. Hence, India today, especially has taken digitalisation to the next level. 

We have the advantage that there are no legacy systems. In simple terms, we kicked off with the cloud, with 2g, 3g, and more updated technologies. So, there is no retrofitting challenge that other mature economies have. Hence, if you could solve a problem for India, you can solve it anywhere across the globe considering the varying languages, the population, and digital literacy levels. 

At Zohocorp, we are working on Indic LLMs which is a challenging task. If we can perfect Indic LLMs, it will be easier to do German LLM or Spanish LLM. Therefore, ManageEngine is bullish on the way India is evolving. Looking at the kind of problems that we have in India like the digital divide, a push like UPI for online payments or ONDC for e-commerce are green flags portraying how we are increasingly attaining digital maturity. Aother major example is the Government of India’s PassportSeva. So, now renewing your passport is a breezy affair with no hassle. 

I say we are on the right track, given our scale and diversity in languages, culture, and more. Hence I say, if we could solve a problem for India, we can solve it anywhere.

In what ways AIOps can enable enterprises to transition to a data-driven, proactive way of functioning leaving behind the legacy approach?

IT, as a department in organisations, has moved to the boardroom and is taken seriously in the post-pandemic era. It is not an afterthought anymore. In the past few years, we have seen a lot of security issues surfacing. One of the reasons is that the sensitive information was only accessible from the office network earlier, while during the pandemic, due to remote work similar information was accessed by employees through their home network. Today, there is more emphasis on security and more power to the attackers because of AI, especially with LLMs in programming. 

Although it is not easy to penetrate an enterprise with all the legacy code, it is easy to generate malware or ransomware through AI. Therefore, the way to go about this is to use AI in the defense as well. If you get 10 failed logins for 10 minutes on a Monday morning around 09:00 am, it is normal. If the same thing happens on a Saturday morning at 03:00 am, then most probably someone is trying to brute force. Therefore, enterprises need to have ML capabilities that will look at past data and try to predict what is normal at a given time. Ransomware and malware can penetrate the system in stealth mode and be a good boy. However, a few days later they start exfiltrating sensitive information to random domains. So an AI deployment can profile your network traffic and spot the sudden transfer of data to a suspicious IP or detect any unusual behaviour and prompt alerts. 

Moreover, cloud deployments are increasingly becoming agile in the present era. Traditionally we would see one release every three months, today, almost every day there are new releases and deployments. However, at the same time, it becomes important to ensure customers’ digital experience should not be impacted while or after the updates. 

Connecting all the processes to improve interoperability and develop a more mature ecosystem, enterprises can leverage a BI tool to dashboard comprehensive data from various departments. A holistic picture will help bring a competitive edge and improve security as all security incidents will be displayed on the dashboard. Now, if you have an AI system that can sift through the data, the next time you raise a firewall configuration change request, the AI tool will automatically prompt actions to avoid any anomaly that would probably have occurred previously if a similar action had been taken.

How has the Indian market witnessed the dynamicity in the adoption of AIOps and spend management for emerging IT solutions over the years?

We are seeing increasing hyperscalar and public cloud spending lately. However, how do you optimise these costs? Be it the dependence on domain expertise versus helping some new intern or a fresher in the team, AI is empowering them to make better decisions and optimise costs. 

A few years from now, IT discussions in the boardroom were in the backseat and it was difficult for CIOs to get budgets sanctioned from their CFOs for IT upgrades and deployments. Thankfully, the scenario has changed to a larger extent. The question today is, can cost be an important factor considering the adoption of AI? So, for MSMEs, the cost of AI adoption can be a dealbreaker. Besides the cost of AI solutions, the cost of operations has gone up as well. Simply, because one would need high-power compute and GPUs to run AI models. 

At ManageEngine, we have not charged for our AI features yet, it comes bundled with the plan. We are not in the immediate scope to make these features chargeable because we aim for more adoption. 

We are taking a staggered approach to AI just to keep the costs in check. We start with narrow models, which are trained on the CPU, and the inference is also done on the CPU. Then we have small language models, trained on GPUs but the inference is done on CPUs. Moreover, we have medium language models trained on lower-powered GPUs. Further, there are emergent capable large language models, which require expensive GPUs to train and to operate. Now with the growing computing, the cost also increases. However, if not today as much, the AI costs will become a key factor in the coming years.

Can you shed some light on a few use cases for Causal AI and how data-driven decision-making has enabled CIOs today to respond better?

Causal AI is a model that analyses cause-effect-based relationships and generates insights for a user to make an informed decision. Consider this scenario, in a complex interconnected IT environment, you did a configuration change and that led to the system going down after around six hours. Now, the two may not be directly connected but one causes the other. The configuration change happened six hours earlier, but the error propagated into the system silently at a sloth’s pace and eventually caused the system to go down after six hours. This is where Causal AI can play a significant role in preventing such a failure. 

At ManageEngine, we have built Causal AI graphs that help the user look at the dependencies between systems and make an informed decision. If an incident has occurred, it is better to identify the root cause of the incident with the causal AI framework than with a correlation-based AI framework. Besides IT, as a department,  it could help in other operational areas as well. For example, if your sales are slumping, what could be the root cause? Although there could be multiple reasons, analysing it through the Causal AI framework could let you know that maybe you ran a social media promotional campaign but due to a low budget, it restricted your access to a larger audience, hence the lesser turnout. 

Therefore, identifying and raising alerts on probable red flags in the system is just one part of Causal AI. The way we position AI to our customers is that it is another tool in your kitty that helps you resolve issues and make effective decisions.

What plans does ManageEngine have for the Indian market in the coming years? Are we expecting any new launches?

We are continuously working in multiple fields and we are bullish on both Causal AI and Agentic AI. Agentic AI is a bigger bridge between LLM and enterprises because in enterprises, you have a lot of structured and semi-structured information and LLMs work well on unstructured information. You need that connect to that structured information in your enterprise database to give real-time access. So these agents will be the missing links to LLMs in the enterprise. So we are working on it. Moreover, we are bullish on India. We are working on Indic LLMs and launching it soon to promote the adoption in India faster and better.