Sean Mullaney is the Chief Technology Officer at Algolia, an end-to-end, AI-powered search and discovery platform.
Sean is a former Stripe and Google executive with a background in scaling engineering organizations, developing AI-powered Search and Discovery tools, and growing API-first solutions globally. At Algolia, he’s overseeing the technology behind the second-largest search engine after Google that’s being used for over 1.5 trillion searches each year. Most recently, he led the company’s launch of AlgoliaNeuralSearch – the world’s fastest, hyper-scalable, and cost effective vector and keyword search API.
What initially attracted you to computer science?
When I was 10 years old, my parents bought our first computer into the home. The very first thing I wanted to do was figure out how to write a text adventure game that I was copying out of a book. A few years later, I started learning C++, but designing and building computer games remained a really big passion of mine as a teenager just beginning to explore computer science.
You spent over 7 years at Google, where you helped to build and lead teams working on strategy, operations, big data and machine learning. What was your favorite project and what did you learn from this experience?
We figured out how to use all the big data we had on how advertisers used our products to help sales teams. We wrote developed custom rules (later more complex neural networks) to predict which customers we should approach with which products at which times to maximize the likelihood of a salesperson’s time resulting in revenue uplift. With over 1 million advertisers on Google, this tool significantly helped the sales teams find the needles in the haystacks.
In a recent DevBit wrap up, you described the purpose of Algolia as being to enable users to index the world and to put content in motion. Could you elaborate on what this statement means?
Ultimately, we want to help our customers get value out of their data. The internet has created such a massive explosion of content and e-commerce products and, while this development is certainly a significant milestone, the sheer overwhelming amount of information now available means that it’s also harder than ever–and becoming increasingly difficult–to find what you are actually looking for as a user. However, when search and discovery is powered by AI, the growing list of content can be intelligently accessed and put into motion to truly help users, not just overwhelm them.
In September 2022, Search.io and its proprietary flagship product NeuralSearch™ was acquired by Algolia, can you discuss what this search technology is specifically?
In a nutshell, Algolia NeuralSearch integrates keyword matching with vector-based natural language processing, powered by LLMs, in a single API – an industry first. The solution incorporates our proprietary and first-of-its-kind Neural Hashing technique that makes the use of vectors scalable and 90% more cost-effective to use – an issue other AI companies, including ChatGPT, face. What’s really exciting about this breakthrough product is that it makes true AI search scalable for enterprise-grade organizations.
The new technology also allows customers, such as retailers, to understand and deliver content that matches queries that are normally too conversational to deliver accurate or any results (considered long-tail). These make up 55% of current site searches. As the only end-to-end AI search solution that applies AI across query understanding, retrieval, and ranking, NeuralSearch truly understands these queries and turns missed opportunities into revenue.
Outside of Neuralsearch™, what are some of the other machine learning methodologies that are used?
We incorporated AI across three primary functions–query understanding, query retrieval, and ranking of results. We at Algolia call this the AI search sandwich:
- Query understanding: Algolia’s advanced natural language understanding (NLU) and AI-driven vector search provide free-form natural language expression understanding and AI-powered query categorization that prepares and structures a query for analysis. Moreover, Adaptive Learning based on user feedback fine-tunes intent understanding.
- Retrieval: The most relevant results are then retrieved and ranked from most to least relevant. The retrieval process merges the Neural Hashing results in parallel with keywords using the same index for easy retrieval and ranking. This approach solves the ‘null results’ problem and significantly improves click positions and click-through rates. No other search platform in the search and discovery space offers this powerful capability.
- Ranking: Finally, the best results are pushed to the top by Algolia’s AI-powered Re-ranking, which takes into account the many signals attached to the search query, (including the exact keyword matching score, the contextual personalization profile, the observed popularity of items, the semantic matching score, etc.) and learns to reach maximum relevance.
Additionally, as the index changes, new products are added, new content is uploaded, or as terms take on new meaning, the AI-powered Algolia NeuralSearch product will learn and adjust automatically. It doesn’t require any additional headcount or manual operations. It will automatically match keywords or concepts—possibly a mix of both—depending on the query or search phrase. This truly puts search on autopilot.
Algolia recently increased its free plan from offering 10000 records, and bumped it up to 1 million records, what was the mindset behind this, and how has the market reacted?
We specifically chose to evolve Algolia’s pricing and packaging to be even more developer-friendly with the introduction of two new developer-oriented plans: a “build” plan that is free and a “Grow” plan that offers easy scalability at affordable prices. The new Build plan increases the number of free records that a developer can store in Algolia from 10,000 to now 1 million records. This represents a 100x increase in the number of free records developers can now index in Algolia. Additionally, Algolia slashed the cost of search requests in its Grow plan by 50% and records by 60%.
The idea behind our updated “Build” pricing plan is to provide developers with free access to the entire set of capabilities in its AI-powered Search and Discovery platform. The “Grow” plan, for when a developer is ready to scale their application, enables developers with more developer-friendly usage-based pricing for live production settings.
One important note here is that any designer, creator, or builder—whether they are a casual or fully committed software engineer—can quickly and easily access all the tools, documentation, sample code, educational content, and cross-platform integration capabilities needed to get started with managing their data, building a search front-end, configuring analytics, and more – all for free. Moreover, they will have immediate access to a growing developer community of more than 5 million builders.
Can you discuss the search personalization tools that are offered?
Algolia offers several search personalization tools for companies to harness data to better improve recommendations, including different kinds of recommendations and unique ways to leverage data to actually drive these recommendations.
A few examples include:
- Trending: Suggest other items that are trending in popularity and related to the searches your customer has performed.
- Ratings-based: People want to buy products with the best ratings.
- Personalized: Based on what you purchased last time, browsing history, location, or other factors, we recommend these other products.
These data-driven methods can help to quickly enhance and improve results based on how customers interact with products, so you’re more likely to recommend the products that actually convert the best.
You’ve described Algolia as being the most scalable hybrid AI search engine in the world. How has Algolia been designed to scale so efficiently?
It all comes back to Neural Hashing. This cutting-edge solution compresses and dramatically speeds up each and every query. It’s much faster to compute hashed similarity than standard vector similarities and returns results in milliseconds.
Neural Hashing represents a breakthrough for putting AI retrieval into production for a huge variety of use cases. Combined with AI-powered query processing and re-ranking, it promises to unleash the full power of AI on-site search. Prior to Algolia’s proprietary breakthrough, vector-based search has been too computationally expensive to run in production.
The part of the sandwich I’d like to focus on most is the meat: retrieval. The reason we say we’re the only true end-to-end AI search engine is because there has been a constant battle behind the scenes in the search industry to add AI to retrieval. Information retrieval is an incredibly complex process, and it’s even more complex to master high-performing, cost-effective AI retrieval at scale. We mastered it with our breakthrough Neural Hashing technique. In doing so, we essentially won the quest for AI search’s Holy Grail.
Is there anything else that you would like to share about Algolia?
It’s an exciting time to be working at Algolia, and we’re always looking to start conversations with talented, passionate people who want to join us on our journey to build the world’s best search technology. If that sounds like you, I’d invite you to check out our current openings at https://www.algolia.com/careers/.
- The Black Box Problem in LLMs: Challenges and Emerging Solutions
- Alex Ratner, CEO & Co-Founder of Snorkel AI – Interview Series
- Circleboom Review: The Best AI-Powered Social Media Tool?
- Stable Video Diffusion: Latent Video Diffusion Models to Large Datasets
- Donny White, CEO & Co-Founder of Satisfi Labs – Interview Series