How Perplexity.ai Build Its AI-Powered Conversational Search Assistant
Discover the AI product design, user-centric principles, data strategies, and key AI/ML technology stack behind Perplexity.ai's search engine and conversational assistant.
Welcome to the AI Product Craft, a newsletter that helps professionals with minimal technical expertise in AI and machine learning excel in AI/ML product management. I publish weekly updates with practical insights to build AI/ML solutions, real-world use cases of successful AI applications, actionable guidance for driving AI/ML products strategy and roadmap.
Subscribe to develop your skills and knowledge in the development and deployment of AI-powered products. Grow an understanding of the fundamentals of AI/ML technology Stack.
Perplexity.ai is an AI-powered search engine and conversational assistant that provides natural language responses to user queries by summarizing and citing relevant information from web sources. It summarizes web information, with a focus on providing credible, up-to-date answers with inline source citations. Perplexity.ai has emerged as a game-changer, offering a unique conversational search experience powered by cutting-edge AI technologies. By seamlessly blending natural language processing, large language models, and web crawling capabilities, Perplexity has redefined the way users interact with and consume information online. This article delves into the intricate details of how Perplexity builds its conversational search product, shedding light on the innovative data strategies and AI technologies that fuel this groundbreaking platform.
How Perplexity.ai Work
Perplexity provides a conversational, personalized search experience by engaging users in dialogue, summarizing relevant web information with sources, maintaining context across sessions, and leveraging advanced AI models to understand and cater to each user's specific intents and needs. here's how Perplexity.ai's conversational search product works from a user perspective:
Natural Language Queries
Users can enter queries in natural language, just like asking a question to another person. Perplexity aims to understand the intent behind the query rather than just matching keywords.Conversational Dialogue
Perplexity engages users in a conversational dialogue by asking clarifying follow-up questions to better understand the context and user's specific information needs. This allows it to provide more relevant and personalized responses.Summarized Answers with Sources
Instead of just providing a list of links, Perplexity generates a concise summarized answer to the query by synthesizing information from multiple web sources. It includes inline citations in brackets to indicate the sources used.Multimedia Responses
In addition to text, Perplexity can include relevant images, charts, or other multimedia in its responses to enhance the answer.Maintaining Context
Features like "Your Threads" allow users to save previous search conversations. Perplexity maintains this context across sessions to provide more relevant follow-up responses.Personalized Experience
Perplexity likely tracks user history, preferences, and interactions to further personalize the search experience over time. The "Copilot" feature leverages GPT-4 to guide users through their queries for a seamless personalized journey.Model Selection (Paid Tier)
For free searches, Perplexity uses GPT-3.5 combined with its own model. The paid "Pro" tier gives access to more advanced models like GPT-4, Claude, and Perplexity's proprietary pplx-70b for enhanced performance.
So in essence, Perplexity provides a conversational, personalized search experience by engaging users in dialogue, summarizing relevant web information with sources, maintaining context across sessions, and leveraging advanced AI models to understand and cater to each user's specific intents and needs.
Key AI/ML Technology Stack Behind Perplexity.ai's Conversational Search Product
Perplexity AI’s platform leverages the power of AI, particularly OpenAI's GPT-4 and Anthropic's Claude 2, to provide instant, up-to-date answers through natural language interaction. It offers a free version that uses its Perplexity model based on OpenAI's GPT-3.5 model, combined with the company's standalone large language model (LLM) that incorporates natural language processing (NLP) capabilities. Additionally, there is a paid version called Perplexity Pro, which has access to GPT-4, Claude 3, Mistral Large, Llama 3.
Perplexity’s proprietary large language models (LLMs)
Perplexity utilizes open-source LLM architectures as a starting point, they have developed their own proprietary pplx-7b-online and pplx-70b-online models fine-tuned and augmented with web data to deliver accurate, timely and helpful responses.
Perplexity.ai's online LLMs are called pplx-7b-online (7 billion parameters) and pplx-70b-online (70 billion parameters).
These models build upon the open-source mistral-7b and llama2-70b base models, respectively.
Perplexity has fine-tuned these models using their own curated training data to optimize for helpfulness, factuality, and providing up-to-date information from the internet.
The models leverage Perplexity's in-house search, indexing and web crawling technology to augment their knowledge with relevant excerpts (snippets) from websites.
This allows Perplexity's LLMs to accurately answer time-sensitive queries by incorporating the latest information beyond just their training data.
Perplexity evaluates their LLM performance on curated datasets testing helpfulness, factuality, and ability to provide fresh/up-to-date information.
Combining Multiple Language Models
Perplexity utilizes an ensemble of large language models like GPT-3.5, GPT-4, Claude, and its own proprietary pplx-7b and pplx-70b models.
For free searches, it primarily uses GPT-3.5 combined with pplx-7b. The paid "Pro" tier provides access to GPT-4 and pplx-70b.
This model ensemble approach leverages the strengths of different LLMs to generate comprehensive and accurate responses, particularly OpenAI's GPT-4 and Anthropic's Claude 2.
Additionally, there is a paid version called Perplexity Pro, which has access to GPT-4, Claude 3, Mistral Large, Llama 3.
Fine-Tuning on Curated Data
Perplexity has fine-tuned its LLMs, like pplx-7b and pplx-70b, on proprietary curated training data optimized for helpfulness, factuality, and up-to-date knowledge.
It evaluates model performance on custom datasets testing for accurate citations, factual responses, and ability to incorporate the latest web information.
Contextual Search with Clarifying Questions
Perplexity engages users in a conversational dialogue, asking clarifying questions to better understand the context and intent behind queries.
This allows it to retrieve the most relevant web excerpts and generate highly personalized, context-aware responses tailored to the user's needs.
Perplexity's technology stack combines large language models fine-tuned on curated data, web crawling and indexing, conversational dialogue capabilities, cloud infrastructure for model training/deployment, and personalization techniques leveraging user data to power its conversational search experience.
AI Product Design and User-Centric Principles in Perplexity.ai’s Conversational Search Product?
Perplexity.ai has implemented several AI product design and user-centric principles in its conversational search product:
1. Conversational Interface and Clarifying Questions
Perplexity engages users in a conversational dialogue, asking clarifying follow-up questions to better understand the context and intent behind queries.
This allows it to retrieve the most relevant information and generate highly personalized responses tailored to each user's specific needs.
2. Transparent Source Citations
Perplexity provides inline citations in brackets for the web sources used to generate its responses, ensuring transparency and factual accuracy.
This builds trust by allowing users to verify the information and trace it back to the original sources.
3. Prioritizing Relevance and User Intent
Rather than simply matching keywords, Perplexity aims to interpret the actual intent behind queries and provide relevant, comprehensive answers.
It deploys long-tail queries and engages in dialogue with users to refine and better understand their needs.
4. Ad-Free, User-Centric Experience
Perplexity provides an ad-free environment that prioritizes the user experience over sponsored content or advertisements.
This allows it to focus solely on delivering the most helpful and relevant information to users.
5. Personalization Features
Features like "Your Threads" allow users to save previous search interactions for easy reference and context across sessions.
The "Copilot" feature leverages GPT-4 to guide users through the search process for a seamless, personalized experience.
6. Modern UI/UX Design
Perplexity has an incredibly user-friendly interface designed to be self-explanatory and reduce friction.
It reflects a commitment to user-centric design principles and delivering a human-like conversational experience.
7. Mobile-First Approach
Perplexity is actively focusing on expanding its mobile presence with a mobile-native search experience incorporating voice interaction and intuitive gestures.
By prioritizing conversation, personalization, transparency, relevance over keyword matching, modern UI/UX, and an ad-free environment, Perplexity has implemented key AI product design and user-centric principles to deliver a differentiated, user-focused conversational search experience.
How Perplexity.ai Use User Data to Personalize Search Results
Perplexity.ai employs several strategies to leverage user data and personalize search results:
Conversational Dialogue and Clarifying Questions
Perplexity engages users in a conversational dialogue, asking clarifying follow-up questions to better understand the context and intent behind their queries.
This allows Perplexity to retrieve the most relevant web excerpts and generate highly personalized responses tailored to each user's specific information needs and preferences.
User History and Preferences
Perplexity likely tracks and analyzes user search history, interactions, and preferences to further personalize results over time.
Features like "Your Threads" allow users to save previous search conversations, which Perplexity can leverage to maintain context across sessions and understand user interests.
User Profile Settings
Users can adjust settings in their "AI Profile" like language, output format, and tone preferences.
As users update their AI Profile, Perplexity becomes better at generating personalized results aligned with their specified preferences.
Combining Multiple Language Models
Perplexity utilizes an ensemble of large language models like GPT-3.5, GPT-4, Claude, and its own pplx-7b and pplx-70b.
This allows it to combine the strengths of different models to generate more comprehensive, context-aware responses personalized to each user.
Fine-Tuning on User Data
Perplexity likely fine-tunes its language models on user data like search histories and interactions to improve personalization.
It evaluates model performance on the ability to incorporate user context and preferences into responses.
So in essence, Perplexity leverages conversational interactions, user history, preferences, profile settings, multiple language models fine-tuned on user data, and context from previous searches to continuously learn and refine its ability to provide accurate, personalized search results tailored to each individual user.
How is Perplexity.ai Impacting the Search Market?
Perplexity is innovating against traditional search engines like Google from a market perspective by offering a conversational AI-powered search experience that provides concise answers with citations directly from the query. By offering a conversational AI interface with citations, focusing on understanding user intent, and leveraging AI for summarization directly from the query, Perplexity is innovating against the traditional search engine model exemplified by Google. This differentiates Perplexity in the market from a user perspective.
Industries Benefiting from Perplexity.ai
A wide range of industries and enterprises can benefit from the capabilities of Perplexity.ai in several ways:
Research and Data Analysis
Perplexity can streamline research processes like market analysis, competitor research, and data analysis by providing comprehensive summaries and insights from diverse sources.
Enterprises can use it to quickly gather detailed information on topics like consumer behavior, market trends, and emerging technologies to make more informed strategic decisions.
Content Creation and Marketing
Content marketers can leverage Perplexity to generate ideas, create content outlines, optimize copy for SEO, and craft engaging marketing materials tailored to different platforms.
It can assist with content ideation, research, and creation for blogs, websites, social media, and more.
Professional Development and Training
Perplexity allows employees to easily research industry trends, best practices, and acquire new skills by summarizing information from various sources.
It can be used for professional training, upskilling, and keeping teams updated on the latest developments in their field.
Recruitment and Talent Sourcing
Recruiters can utilize Perplexity for in-depth candidate research, analyzing industry trends, crafting job descriptions, and improving overall recruitment strategies.
It serves as a research assistant to gather real-time information on candidates, skills, salaries, and competitors.
Customer Service and Sales
Enterprises can use Perplexity to provide accurate and efficient responses to customer inquiries, enhancing customer experience.
Sales teams can leverage it to research prospects, craft compelling pitches, and expedite the sales process.
Investment Analysis and Due Diligence
Financial firms and investors can employ Perplexity for thorough investment analysis, researching companies, industries, and simulating scenarios.
Other industries that can benefit include healthcare, legal, journalism, academia, and more by using Perplexity for research, content creation, analysis, and staying updated on the latest information in their respective fields.
Conclusion
Perplexity.ai's conversational search engine represents a significant leap forward in the world of search, combining the power of AI technologies with innovative data strategies to deliver a truly personalized and context-aware search experience. By leveraging web crawling, multiple language models, curated training data, conversational interactions, and user data, Perplexity has set a new standard for how users discover and consume information online. As the demand for more intelligent and intuitive search solutions continues to grow, Perplexity's pioneering approach positions it as a frontrunner in shaping the future of search.