The recent advancements in LLMs improved their performance and made them more affordable – this unlocked multiple possibilities for companies to integrate LLMs into their products. Indeed, there have been a lot of impressive demos. But how do companies actually use LLMs in production?
We put together and regularly update a database of 450 use cases from 100+ companies that detail real-world applications and insights from ML and LLM system design. In this blog, we share 20 selected examples of LLM-powered products from various industries.
The database is maintained by the team behind Evidently, an open-source tool for LLM and ML evaluation and observability. Give us a star on GitHub to support the project!
👷 LinkedIn extracts skill information from texts
They extract skills from various content across the platform and map these skills to their Skills Graph to ensure accurate job and learning matches.
🗝 Google speeds up security and privacy incidents workflows
They use LLMs to summarize incidents for different audiences, including executives, leads, and partner teams. It saves responders’ time and improves the quality of incident summaries.
🏪 Picnic improves search relevance for product listings
They leverage LLMs to enhance product and recipe search retrieval for users from three countries with their own unique language and culinary preferences.
🙅 Yelp detects inappropriate language in reviews
The company enhanced its content moderation system with LLMs to help identify egregious instances of threats, harassment, lewdness, personal attacks, or hate speech.
🚗 Uber tests mobile applications
They created DragonCrawl, a system that uses LLMs to execute mobile tests with the intuition of a human. It saves thousands of developer hours and reduces test maintenance costs.
#️⃣ Grab automatically tags sensitive data
They use LLM to classify data entities, identify sensitive data, and assign the most appropriate tag to each entity.
🛒 Instacart builds an internal AI assistant
Teams use an internal AI assistant called Ava to write, review and debug code, improve communications, and build AI-enabled internal tools on top of the company’s APIs.
🛍 Whatnot detects marketplace spam
They use LLMs to enhance trust and safety areas like multimodal content moderation, fulfillment, bidding irregularities, and general fraud protection.
💌 Nextdoor generates engaging email subject lines
The company aims to generate informative and engaging subject lines that will lead to more email opens, clicks, and eventually more sessions on the platform.
🍿 Vimeo builds customer support AI assistant
They prototyped a help desk chatbot where customers input their questions and receive immediate, accurate, and personalized responses.
🤖 GoDaddy classifies support inquiries
GoDaddy leverages LLMs to enhance customer experience in their messaging channels by classifying support inquiries. They share lessons learned operationalizing these models.
🗞 OLX extracts information from job listings
They use Prosus AI Assistant, their generative AI (GenAI) model, to extract job roles in job ads and ensure a closer alignment between job seekers’ desired jobs and the relevant listings.
🔢 Honeycomb helps users write data queries
The company built Query Assistant to accelerate users’ learning curve associated with queries. Users can describe or ask things in plain English like “slow endpoints by status code” and Query Assistant will generate a relevant Honeycomb query to iterate on.
📦 DoorDash extracts product attributes from unstructured SKU data
They use LLMs to extract and tag product attributes from raw merchant data. It allows to easily match customer queries with relevant items on DoorDash and helps delivery drivers to find the correct product in the store.
⚠️ Incident.io generates summaries of software incidents
Incident.io helps to collaborate on software incidents by suggesting and updating the incident summary. This suggestion considers the latest update, the conversation in the Slack channel, and the previous summary. Half of all summary updates in Incident.io are now written by AI.
🪡 StitchFix generates ad headlines and product descriptions
The company combines algo-generated text with a human expert-in-the-loop approach to streamline crafting engaging advertisement headlines and producing high-fidelity product descriptions.
💳 Digits suggests questions about banking transactions
They use generative models to assist their customers – accountants – by suggesting questions about a transaction to a client. The accountants can then send the question to their client as is, or edit it without typing everything from scratch.
🧑🏫 Duolingo generates content for lessons
The company leverages LLMs to help their learning designers come up with relevant exercises for lessons. Human experts plan out the theme, grammar, vocabulary, and exercise types for a given lesson and the model outputs relevant exercises.
🏠 Zillow detects discriminatory content in real-estate listings
The company uses LLMs to understand whether real-estate listings contain proxy for race and other remnants of historical inequalities in the real estate domain.
🍲 Swiggy improves search relevance in hyperlocal food delivery
They use LLMs to match search queries in a variety of languages with millions of dish names with regional variety.
Want more examples of LLM systems in production?
Check out our database of 450 use cases from 100+ companies that share their learnings from implementing ML and LLM systems. Bookmark the list and enjoy the reading!
Top comments (3)
LLM and ML just are TLM (Two/Three Letter Acronym) and should probably be spelled out early to give context to the reader. I needed to go look elsewhere.
The section header links present in the browser like advertisements but aren't. It would be helpful to connect the content with the section if the content didn't refer to they, but rather use the subject from the section header.
Very valuable resources for everybody. Good work!
Very nice