“Our first priority is to ensure a seamless and uninterrupted streaming experience”

Pramod Prakash, CTO, Zee Entertainment Enterprises, explains how the company combines cloud-based AI services with in-house engineering, manages streaming workloads during major live events, monitors recommendation models and uses automation to strengthen content protection.

Pramod Prakash, Chief Technology Officer, Zee Entertainment Enterprises

As streaming platforms scale their content libraries, user bases and live-event offerings, the technology challenge is no longer limited to delivering uninterrupted video. Platforms must balance unpredictable viewing demand with increasingly intensive AI workloads, maintain recommendation quality, safeguard viewer data and protect premium content against piracy. These priorities require an infrastructure strategy that combines scalability with control over critical systems and workflows.

At Zee Entertainment Enterprises, the approach brings together cloud-hosted capabilities, managed AI services and model APIs, alongside investments in in-house infrastructure. While external AI services support content analysis and the identification of engaging moments, the company retains control over its backend pipeline, content management integration, publishing workflows and frontend experience. It is also working towards hosting open-source large language models on its own GPU infrastructure.

In this conversation with CIO&Leader, Pramod Prakash, Chief Technology Officer, Zee Entertainment Enterprises, discusses how the company is approaching AI infrastructure, separating streaming and AI workloads during major live events, monitoring recommendation systems for performance degradation, strengthening data-protection practices and deploying automated safeguards against content piracy.

CIO&Leader: How would you describe ZEEL’s current AI estate? What is being consumed through model APIs, what is embedded in SaaS platforms or managed services, and what is being developed and run in-house across cloud and owned infrastructure?

Pramod Prakash: Our current AI estate brings together cloud-hosted capabilities, managed AI services and model APIs, alongside investments in building our own AI infrastructure. Our recommendation engine, for instance, is developed and hosted on our cloud infrastructure, giving us greater control over the models, data and personalisation capabilities. For Clips, we leverage Gemini through SaaS APIs for AI-powered content analysis and identification of engaging moments.

“We are working towards hosting open-source LLMs on our own GPU infrastructure.”

Looking ahead, we are working towards hosting open-source LLMs on our own GPU infrastructure. This will give us greater flexibility and control over model deployment, cost optimisation and customisation as we continue to develop and scale different AI use cases.

CIO&Leader: In the Clips pipeline, how do you decide which cloud AI services to rent and which capabilities to build or run internally?

Pramod Prakash: For Clips, we leverage Google Cloud’s AI capabilities through SaaS APIs, including Gemini models, Chirp for speech-to-text, Cloud Vision API and deep learning models for image processing. These help us identify engaging moments from movies and shows and transform them into short, vertical videos.

“The entire backend pipeline, CMS integration, publishing workflows and frontend experience have been developed in-house by our engineering teams.”

At the same time, the entire backend pipeline, CMS integration, publishing workflows and frontend experience have been developed in-house by our engineering teams. This approach allows us to leverage specialised AI capabilities while retaining control over how they are integrated into our content workflows and, ultimately, how the experience is delivered to viewers.

CIO&Leader: During a major live event or sports tournament, how do you allocate capacity between streaming demand and AI workloads? What metrics matter beyond uptime?

Pramod Prakash: During major live events, our first priority is to ensure a seamless and uninterrupted streaming experience. We plan and provision capacity based on anticipated viewing demand, while maintaining adequate headroom to accommodate sudden traffic spikes. We also ensure that AI and other non-critical workloads do not compete with the resources required for live streaming.

Where possible, we isolate workloads and use independent scaling strategies, allowing AI processing to be scaled or deferred without affecting the core streaming platform. Beyond uptime, we closely monitor concurrent viewers, playback success rates, video start time, buffering ratio, playback failures, API latency and overall system performance. Ultimately, these metrics help us assess whether viewers can access content seamlessly and remain engaged throughout the event.

CIO&Leader: How do you monitor recommendation and discovery models for degradation? How would you detect if an external model provider changed its model overnight and affected performance?

Pramod Prakash: We monitor recommendation quality through a combination of user engagement and technical performance metrics. On the engagement side, we look at click-through rates, watch time, content consumption, completion rates and how users interact with recommended titles. Alongside this, we track technical indicators such as latency, error rates and overall service performance.

“The objective is to identify performance changes early and maintain consistent recommendation quality and a seamless viewer experience.”

These metrics are continuously compared against established baselines, with alerts configured to flag significant deviations. When we identify an anomaly, it triggers a deeper investigation to understand the underlying cause and, where necessary, make model or system-level improvements. The objective is to identify performance changes early and maintain consistent recommendation quality and a seamless viewer experience.

CIO&Leader: How are you approaching the DPDP Act and India’s evolving data-protection requirements, particularly when using viewer data for personalisation or sharing it with external AI models?

Pramod Prakash: Our approach to data protection is centred on building privacy and security into our technology and product design from the outset. When it comes to personalisation, we focus on using viewer data responsibly, with appropriate consent mechanisms, access controls and safeguards governing how data is collected, processed and used.

Similarly, when working with external AI models and services, we are mindful of the information being shared, ensuring that only necessary data is exposed and appropriate security and data-protection safeguards are in place. As the DPDP framework evolves, we continue to strengthen our processes and technology controls to align with applicable requirements, while ensuring that personalisation and viewer privacy remain equally important considerations.

CIO&Leader: How are you using automation to tackle piracy and protect content? Where do you draw the line between automated action and human intervention to avoid false positives that could block legitimate viewers?

Pramod Prakash: We use multiple layers of protection to address piracy and content abuse, combining automated security mechanisms with appropriate safeguards. These include DRM-based content protection and CDN-level controls that help detect and prevent the misuse of HMAC tokens, particularly unauthorised sharing across multiple users.

We have also implemented additional safeguards for L3 devices, which have inherent DRM security limitations, to minimise the risk of content piracy. Our approach is to ensure that these automated protection mechanisms are supported by appropriate controls to minimise false positives, so that legitimate viewers are not unnecessarily affected while we continue to protect our content.

Share on