
Transforming Telco: AI in Telecommunications
Based on our previous work with telcos and our research, we have identified many impactful AI-led use cases.
In this interview, our Head of Cloud & Partner, Artem Kobrin, talks to Roman Jasins, Co-Founder of Peak Defence, about their journey in automating cybersecurity processes with AI.
In the previous session, we discussed the business impact of AI on security and audit. Now, Artem and Roman discuss the specifics of the solution implementation process and the technologies used.
Since its inception a couple of decades ago, Peak Defence has been focused on helping companies protect their assets and manage risks. Now, they are an AI-powered security guardian angel, securing clients assets.
While Peak Defence has consistently strived to scale the security expertise they have cultivated over the years, scaling anything that necessitates extensive knowledge and human resources was a formidable challenge till recently.
Security has historically been viewed as a disabler, something that closes doors and opportunities, depending heavily on a few skilled, experienced individuals who do a lot of legwork.
But Peak Defence looks at it differently. They were determined to figure out how to achieve an optimal level of security with minimal investment of time and resources and better serve their clients, both start-ups and bigger enterprises.
When AI started to gain traction in the last couple of years, Peak Defence thought, “Why don’t we take AI and see if we can build a prototype where the machines would do all the heavy lifting in terms of context understanding?”
Then, businesses could evaluate the most important information and make informed decisions about what they need to protect. This was when they reached out to Neurons Lab to bring in niche AI expertise to work on the prototype.
For the Peak Defence AI platform, we used Antropic’s Claude model (first Claude2, then the newly-released Claude3) to experiment and save time otherwise needed to gather data and train the model. When selecting the model, there were four key criteria that impacted the decision:
While Peak Defence had a couple of decades of experience in the security space, its data was not cataloged in a useful way. Another challenge was the usability of the data due to privacy issues.
To overcome this cold-start problem, we had to hand-pick data and run numerous tests to see how the model behaves and how it performs. This is an ongoing process.
Now, we have a machine-human hybrid, where AI does a lot of the heavy lifting for the Peak Defence team, and professionals review it, both to deliver quality and to help the model learn.
For the Peak Defence AI platform, the feedback loop and reinforced learning were extremely important. That’s how we get better results for each particular user and also continuously improve the product. The feedback loop allows the model to generate more personalized results and achieve higher relevancy.
Claude 2 supports hundreds of languages out of the box, so that’s why we didn’t need to do any language-specific model training. We also used Amazon Bedrock to transform the documentation information into a digital vector space, which is also multilingual. For example, even if a question was asked in English, the model would generate an answer in the original language from the context.
With any new product, any new problem you’re trying to solve, and especially with AI, there is a lot of experimentation needed because everything is still new and is changing rapidly.
There is no guarantee; all the experiments might fail. So, if you don’t have the right team that is ready to experiment and pivot based on the results, discarding whatever didn’t work quickly, you don’t stand a chance.
For this project, we used a well-tested combination of methodologies, moving in two-week intervals, sprints. The idea was to see if it could be done and learn from that. That’s exactly what helped us in this partnership with Peak Defence and allowed us to create a solution that already sees some traction with customers and an increasing demand after only 3 months into the development.
Firstly, creating a quick prototype to check the feasibility of the idea helped us get a head start. Doing small experiments every day also allowed us to move quickly and find the right setup.
The next step was to create a product that would bring real value. For this, we needed to do some data processing, find the right model, find the database and infrastructure that can handle all of this.
We also needed to address the security, data privacy, and scalability issues. That’s why we made this platform serverless, with most of the heavy lifting done by AWS, which hosts instances and provides serverless services that can be easily integrated through the API.
Here is what the solution looks like now:
The next step would be to simplify this architecture within LangGraph further, putting decision-making and separation on the graph. That’s something we are currently experimenting with.
Having many Lambda functions triggered in parallel speeds up the work of information security professionals and allows us to visualize what is going on in a particular business and what assets need to be protected.
The major benefit of an AI-powered approach is the ability to process a lot of data and get meaningful results with actionable recommendations without having to wait for hours or days. This is one of the things that Peak Defence clients value the most.
The combination of clear visibility and rapid analysis is a powerful capability that Peak Defense provides. It allows information security teams to be much more agile and responsive, which is crucial in today’s fast-paced threat landscape.
We are excited about continuing to build on this foundation and exploring new techniques like LangGraph to enhance the platform further, pushing the boundaries of what’s possible with AI in this domain.
Neurons Lab delivers AI transformation services to guide enterprises into the new era of AI. Our approach covers the complete AI spectrum, combining leadership alignment with technology integration to deliver measurable outcomes.
As an AWS Advanced Partner and GenAI competency holder, we have successfully delivered tailored AI solutions to over 100 clients, including Fortune 500 companies and governmental organizations.
Based on our previous work with telcos and our research, we have identified many impactful AI-led use cases.
We explore advanced attack techniques against LLMs, then explain how to mitigate these risks using external AI guardrails and safety procedures.
We cover some of the most common potential types of attacks on LLMs, explaining how to mitigate the risks with security measures and safety-first principles.
The recently released SWARM framework offers a simple yet powerful solution for creating an agent orchestration layer. Here is a telco industry example.
Traditional chatbots don't work due to their factual inconsistency and basic conversational skills. Our co-founder and CTO Alex Honchar explains how we use AI agent architecture instead.