Launch HN: Needl (YC S22) – Simple search across all your apps
23 by MaxKeenan | 20 comments on Hacker News.
Hi HN, we’re Max and James, co-founders of Needl ( https://www.needl.tech ). Needl is a single search bar for all your apps that lets you find any document, message, or file you need. Here’s a demo video: https://ift.tt/S4V1Cz9 . We’ve all struggled to find information and couldn’t remember if it was shared over Gmail or Slack or Notion. Most of us have hundreds of thousands of emails, documents, messages, etc., and only a small fraction are actually useful. Internal search is even harder than web search in some ways: with web search, there are dozens of right answers and millions of previous searches to train ranking models, but internally (i.e. on your own data), there is only one right answer and limited training data. Needl’s value prop is twofold: it unifies all of your information together and provides better search than existing platforms (here’s an example of Needl compared to Gmail: https://ift.tt/hXcjyNY ). Together, these save time by giving you exactly what you need. Recently we also launched a personal assistant that gives direct answers to any search. Simply ask, “What’s the budget for our 2023 team offsite?” or “When’s my next flight to SF and what’s the confirmation #?” From your web apps, we’ll generate the exact answer you need. (pro tip, you don’t even have to phrase it as a question). To rank results on Needl, we combine text matching, semantic search, and platform-specific usage patterns (e.g. email opens). Semantic search (based on user intent / meaning) has helped us solve the hard problem of internal search – everyone has a different vernacular and way of organization. We use LLMs to create embeddings and then use vector search to allow greater breadth of results over traditional text-based search. Our search (and generative answers) even works well in different languages, something that surprised us. To give generated answers to questions, we take the first results on Needl and run them through GPT-3 and use an engineered version of the user’s query to prompt GPT-3 accordingly. It’s similar to what Phind.com (Hello Cognition) and Perplexity.ai do with web search. Using LLMs to build relationships between documents for search and recommendation was a remnant of our failed college startup. We had built a “TikTok for blog posts” that recommended user-written content based on their interactions with other posts. We quickly found out social media moderation was a nightmare (who would’ve thought) and few people enjoy reading any more. It wasn’t until we threw in the white flag and went to our full-time jobs — James as a SWE at Microsoft and me in investment banking at Moelis — that we realized that recommendation + search was the real product. At my job in particular, intranet search was unusable. The primary way people “searched” for information was by sending company-wide e-mail blasts asking if anyone had materials about an industry or company! This got us wondering why internal search sucked so badly and led us down the path of building Needl. Obviously with a product like this, privacy is the #1 concern for users. Some companies have solved this concern by going local first, but that involves significant reduction in quality of search, because LLMs can’t be run locally. We went the opposite route: users’ entire index is stored on the cloud. We know that’s not for everyone, but the results it enables are significantly more powerful, and we make clear commitments to our users: (1) we never access users’ personal information without explicit permission; (2) all information is encrypted both in transit and at rest, and (3) we will never sell users’ information. We’ve gotten SOC 2 Type II compliant to give our users assurances about the way their data is being handled and our dedication to information security. Setting up Needl takes less than 3 minutes—users create an account, grant read access to the platforms they want to search through (Slack, Notion, GSuite, etc.), and we index their information. Once completed, you have instant, cross-platform search in a Spotlight-esque format—control + space opens Needl from anywhere. In terms of pricing, Needl is free for up to 3 integrations and $10/month for unlimited integrations. To learn more, you can download and set up Needl at https://www.needl.tech/ . If you’ve already tried tools in this space, we’d love to hear your experience and what you care about most. We look forward to everyone’s comments!
23 by MaxKeenan | 20 comments on Hacker News.
Hi HN, we’re Max and James, co-founders of Needl ( https://www.needl.tech ). Needl is a single search bar for all your apps that lets you find any document, message, or file you need. Here’s a demo video: https://ift.tt/S4V1Cz9 . We’ve all struggled to find information and couldn’t remember if it was shared over Gmail or Slack or Notion. Most of us have hundreds of thousands of emails, documents, messages, etc., and only a small fraction are actually useful. Internal search is even harder than web search in some ways: with web search, there are dozens of right answers and millions of previous searches to train ranking models, but internally (i.e. on your own data), there is only one right answer and limited training data. Needl’s value prop is twofold: it unifies all of your information together and provides better search than existing platforms (here’s an example of Needl compared to Gmail: https://ift.tt/hXcjyNY ). Together, these save time by giving you exactly what you need. Recently we also launched a personal assistant that gives direct answers to any search. Simply ask, “What’s the budget for our 2023 team offsite?” or “When’s my next flight to SF and what’s the confirmation #?” From your web apps, we’ll generate the exact answer you need. (pro tip, you don’t even have to phrase it as a question). To rank results on Needl, we combine text matching, semantic search, and platform-specific usage patterns (e.g. email opens). Semantic search (based on user intent / meaning) has helped us solve the hard problem of internal search – everyone has a different vernacular and way of organization. We use LLMs to create embeddings and then use vector search to allow greater breadth of results over traditional text-based search. Our search (and generative answers) even works well in different languages, something that surprised us. To give generated answers to questions, we take the first results on Needl and run them through GPT-3 and use an engineered version of the user’s query to prompt GPT-3 accordingly. It’s similar to what Phind.com (Hello Cognition) and Perplexity.ai do with web search. Using LLMs to build relationships between documents for search and recommendation was a remnant of our failed college startup. We had built a “TikTok for blog posts” that recommended user-written content based on their interactions with other posts. We quickly found out social media moderation was a nightmare (who would’ve thought) and few people enjoy reading any more. It wasn’t until we threw in the white flag and went to our full-time jobs — James as a SWE at Microsoft and me in investment banking at Moelis — that we realized that recommendation + search was the real product. At my job in particular, intranet search was unusable. The primary way people “searched” for information was by sending company-wide e-mail blasts asking if anyone had materials about an industry or company! This got us wondering why internal search sucked so badly and led us down the path of building Needl. Obviously with a product like this, privacy is the #1 concern for users. Some companies have solved this concern by going local first, but that involves significant reduction in quality of search, because LLMs can’t be run locally. We went the opposite route: users’ entire index is stored on the cloud. We know that’s not for everyone, but the results it enables are significantly more powerful, and we make clear commitments to our users: (1) we never access users’ personal information without explicit permission; (2) all information is encrypted both in transit and at rest, and (3) we will never sell users’ information. We’ve gotten SOC 2 Type II compliant to give our users assurances about the way their data is being handled and our dedication to information security. Setting up Needl takes less than 3 minutes—users create an account, grant read access to the platforms they want to search through (Slack, Notion, GSuite, etc.), and we index their information. Once completed, you have instant, cross-platform search in a Spotlight-esque format—control + space opens Needl from anywhere. In terms of pricing, Needl is free for up to 3 integrations and $10/month for unlimited integrations. To learn more, you can download and set up Needl at https://www.needl.tech/ . If you’ve already tried tools in this space, we’d love to hear your experience and what you care about most. We look forward to everyone’s comments!
Comments
Post a Comment