Hailing from a humble background in Goa, Amey Porobo Dharwadker was one of the first ML engineers in video recommendations at Meta. Dharwadker has been with the company for over nine years now, serving as the engineering leader at Meta’s Menlo Park office.
The NIT Tiruchirappalli alumni completed his master’s in electrical engineering from Columbia University in NYC with specialisation in computer vision and machine learning.
During his tenure, he authored several research papers on the same, including ‘PIE: Personalised Interest Exploration for Large-Scale Recommender Systems’ and ‘CAM2: Conformity-Aware Multi-Task Ranking Model for Large-Scale Recommender Systems’. The goal of these papers was to advocate for industry policies and mould the broader AI landscape with fairness and user-centric principles.
His papers were presented at the prestigious ACM web conference.
At present, his works include video personalisation but through responsible recommendations. “My work involves a deeper understanding of user preferences and satisfaction, tackling various biases inherent in ML-based ranking algorithms, and facilitating connections between users and high-quality, interest-relevant content,” Dharwadker told AIM.
Making Facebook Ads More Personalised with AI
Meta relies heavily on advertising, particularly from Facebook and Instagram, constituting a significant portion of its revenue. Analysts predict Meta’s advertising revenue to reach $123.7 billion in 2024, with a large chunk (55.8%) expected to be generated from Facebook ads.
In 2022, the company’s total revenue amounted to $116.61 billion, with advertising contributing approximately 98%.
Over the past decade, Dharwadker’s research areas at Meta have spanned the domains of video recommender systems and user personalisation modelling.
“I developed and deployed large-scale ranking models using deep neural modelling and multi-task learning. These models significantly contributed to the success of products like Facebook Videos, News Feed, and Ads,” Dharwadker told AIM.
However, with the inclusion of AI, the ad space has been changing as well. For example, AI-powered personalised ad targeting has revamped advertising effectiveness. In contrast to traditional methods that relied on demographics and basic online behaviour, AI-driven segmentation delves into extensive data, analysing behavioural, social media, demographic, and textual information.
AI algorithms identify hidden connections and create highly specific micro-segments, offering predictive insights into future user behaviour.
Even though Meta never disclosed if the company was using its Llama 2 family of open-source LLMs for personalisation and recommendation, Dharwadker opines that the burgeoning emergence of multimodal LLMs is instrumental in shaping the future of recommendations. He said that these models can mirror the way humans interact with information and generate content by integrating multiple modalities.
Dharwadker further said that by analysing content and understanding user preferences deeply, these algorithms provide more relevant recommendations, enhancing user satisfaction and engagement. Meta AI’s Reels, for example, assist in decision-making for places to visit, offer tutorial videos for learning new dance moves, and provides inspiration for various projects.
“Simultaneously, the rise of AI democratisation is shaping the landscape with advanced machine learning tools becoming more accessible, making AI user-friendly,” he added, emphasising that the convergence of real-world multimodal applications and AI democratisation holds promise for addressing societal challenges.
The future of generative AI in video recommendations holds promising opportunities for innovation and growth. “As algorithms evolve and data expands, we can anticipate advancements in content creation, enhanced recommendation accuracy, and improved user experiences,” he said.
Working at Meta
Meta is known to foster an inclusive and collaborative work culture. “The vibrant work culture, autonomy to make a positive impact on human lives and the opportunity to work alongside brilliant minds advancing cutting-edge technology were lucrative,” said Dharwadker.
He added that the engineering-centric environment fosters a sense of ownership, allowing for the development and application of innovative solutions to tackle real-world challenges. The culture at Meta values deep technical contributions.
Looking ahead, Dharwadker believes, “This year, we’ll see more examples of trustworthy and responsible AI systems seamlessly embedding intelligence into our everyday lives, making everything from healthcare to transportation to entertainment smarter, more efficient, and ultimately, more human-centric.”