Protecting customer data is foremost priority: ShareChat & Moj's Mukherjee

Mohalla Tech, the parent company of short video platform Moj and social media platform ShareChat, in its recent fundraise of $266 million in Series G almost doubled its valuation to $3.7 billion. This shows the increasing uptake of social and video sharing platforms among users. Moj has a monthly active user base of 160 million and using tech to keep this user base to its platform has been crucial for growth. Deepsekhar Choudhary in an email interview with Debdoot Mukherjee, vice-president AI and Data Science, ShareChat & Moj, talks about how tech is being used to grow the video platform and democratise content creation. Edited excerpts…

Do you think Indian social media platforms have as good a recommendation engine as the likes of Facebook and TikTok?

Our teams are definitely pushing the boundaries of the state-of-the-art recommender systems. We have the largest AI team in India ecosystem, spread across US, Europe and India.  

ShareChat was one of the first apps to take an AI-first approach to create social media feeds, one that we have fine-tuned over the years. From the outset, we did not adopt the conventional “follow-driven” approach where a user’s feed can only show posts which have been acted upon by their social connections. Instead, we designed our feeds to recommend posts which are deemed to be relevant entirely based on the user’s behaviour (e.g., likes, shares, watches) on the platform; thus delinking their experience from the activity of their connections.

With the AI-first approach for content recommendation, we have also expanded the reach for our creator community. In a follow-driven social network, a creator’s reach is limited to their followers and building followership takes time — especially if you aren’t a celebrity. We expect our creators to focus on making high-quality content. Our AI engines do the matchmaking of their content to the relevant users, helping them increase their reach and popularity on our platforms. 

In what ways is ShareChat using AI?

We at ShareChat and Moj use AI to power our recommendation engine, to ensure that the content served on every user’s feed is relevant to their interests and intent. Our platforms utilise hundreds of machine learning (ML) models to deliver a personalised feed to every user based on their interactions on the platform. 

Matchmaking of the right content to the right user enables our content creators to reach the desired success and popularity on our platforms. In order to do this, we have developed cutting-edge AI models that understand and interpret any content posted on our platform in the same way as our users would. These models process visual, audio and text data in order to decode the post’s meaning and serve them to the right user based on their interest. 

Apart from our recommendation engine, creation tools on our apps such as the camera and the video editor heavily employ capabilities in Computer Vision and Augmented Reality.

Moj's first version was built very quickly last year. What kind of major iterations have been done since then and how much did those changes help increase engagement, user numbers?

While building Moj, a lot of technical and product learnings were borrowed from ShareChat. Due to the existing know-how around social and video with ShareChat, a team of over 70 engineers and product experts was formed with a single focus, of building Moj in the earliest possible timeline. Backed with immense dedication, hard work and 30 sleepless hours, the team delivered the first version of the product.

Moj was launched on July 1, 2020, with an aim to get good traction. All the key pillars of Moj have seen significant enhancements since the launch. We have observed steep jumps in user retention on the platform with multiple iterations of the algorithm for feed personalisation, where we have focused on decoding the interests of the user in real-time to readily serve them more relevant content.

We have significantly enhanced our video transcoding algorithms to deliver higher quality videos at lower data bandwidth. The Moj camera has seen major improvements in capturing quality and adding a bunch of filters and effects, which has led to a sharp increase in the volume as well as the diversity of content created on the platform.

Today, Moj has the highest number of monthly active users—160 million amongst Indian short video apps. The Moj user community is spending 34 minutes on an average on the platform daily, this speaks volumes about the app’s stickiness and popularity. 

Data privacy and cybersecurity have emerged as big challenges in tech. What are you doing on this front?

ShareChat is today the largest Indian social media platform and we have a massive user base from different geographies of India. Protecting customers’ data is our foremost priority along with providing an excellent social experience on our platforms. Our Information Security team continually assesses ShareChat based on industry security standards, best practices and regulations and keeps on improving the overall security and privacy posture. Also, following the “security and privacy by default” design principle during the product engineering process helps us make our product and technology more secure. 

Can you highlight some interesting challenges in content moderation and how you solved them with tech?

Every piece of content uploaded on the platform passes through AI/ML based tools that review content. These tools provide various signals and influence our moderation pipeline and have been absolutely essential and instrumental in ensuring that the platform’s health is maintained. 

AI models help us flag content that potentially violates the integrity of our platforms and affects the safety of our users. Millions of posts are created on ShareChat and Moj every day, so it is imperative to employ precise automated methods to screen the content and detect policy violations in a timely manner. We have a wide array of models, each of which specialises in detection of a specific kind of integrity violation such as nudity, violence and spam etc.

How have you been solving the problem of low internet bandwidth and low camera quality in smartphones of users in Tier 2 and 3 cities?

We understand that most of our users have a limited amount of data— that is, between 1 and 1.5GB, which they have to ration across multiple apps. Moreover, most of our users’ devices are old, operating on typically patchy network conditions leading to major drops.

We have worked heavily on our video encoding tech, image compression tech, as well as our CDNs for content delivery to offer the best possible social experience. 

A key challenge is to make our content creation tools — camera, editing, etc work really well on low end devices. These systems are compute-intensive and require a lot of RAM, CPU utilisation and battery. We have worked on optimising our AI models used in our camera, and also optimised our video encoding process so that editing can happen even on old or low-end devices. This is an ongoing process and we would like to enhance our capabilities on this. This is one area that we want to constantly work on and improve further so that we can truly democratise content creation.