Topics
late
AI
Amazon
Image Credits:Kapa
Apps
Biotech & Health
Climate
Image Credits:Kapa
Cloud Computing
Commerce
Crypto
Image Credits:Kapa.ai
initiative
EVs
Fintech
fundraise
Gadgets
gage
Government & Policy
Hardware
layoff
Media & Entertainment
Meta
Microsoft
privateness
Robotics
Security
Social
place
Startups
TikTok
Transportation
Venture
More from TechCrunch
Events
Startup Battlefield
StrictlyVC
Podcasts
Videos
Partner Content
TechCrunch Brand Studio
Crunchboard
Contact Us
YC-alum turns customers’ technical documentation into intelligent assistants
Generative AI and big language models ( LLMs ) have been all the cult in recent eld , upending traditional online search viathe ilk of ChatGPTwhile improvingcustomer bread and butter , content coevals , translation , and more . Now one fledgling inauguration is using LLMs to build up AI assistants adequate to specifically of answering complex questions for developer , software system end drug user , and employees — it ’s like ChatGPT , but for technical merchandise .
establish in February last twelvemonth , Kapa.aiis a alumnus of Y Combinator ’s ( YC)Summer 2023 programme , and it has already accumulate a fairly telling roster of client , including ChatGPT - maker OpenAI , Docker , Reddit , Monday.com , and Mapbox . Not bad for an 18 - month - previous job .
“ Our initial concept came after several friends who ran tech company accomplish out with the same trouble , and after we work up the first prototype of Kapa.ai to address this for them , we landed our first paid pilot within a week , ” chief executive officer and co - founderEmil Sorensentold TechCrunch . “ This led to organic growth through intelligence of mouth — our customers became our biggest advocates . ”
To build on that early traction , Kapa.ai has now raised $ 3.2 million in a seeded player daily round of funding led byInitialized Capital .
Getting technical
In the broadest terms , troupe feed their technical documentation into Kapa.ai , which then serves up an interface where developer and terminate drug user can involve questions . Docker , for example , late launcheda unexampled certification assistant called Docker Docs AI , which provides instant responses to Docker - related questions from within its documentation pages — this is built using Kapa.ai .
But Kapa.ai can be used for myriad exercise cases such as client living , community of interests engagement , and as a workplace supporter to help employee query their company ’s cognition base .
Under the hood , Kapa.ai is based on several LLMs from unlike providers and leans on a motorcar learning framework called retrieval augmented generation ( RAG ) , which enhances the performance of LLMs by enabling them to easy draw from relevant external data sources to provide richer responses .
Join us at TechCrunch Sessions: AI
Exhibit at TechCrunch Sessions: AI
“ We ’re model - agnostical — we wreak with multiple providers , let in using our own models , to employ the best - performing stack and retrieval technique for each specific manipulation case , ” Sorensen said .
It ’s deserving notice that there are a number of like tools out there already , include venture - backed startups such asSanaandKore.ai , which are substantively about bring colloquial AI to enterprise knowledge base . Kapa.ai , for its part , tally into that bucketful , but the company says its main differentiator is that it for the most part focalize on external users rather than employee — and that has had a gravid influence on its invention .
“ When deploy an AI assistant outwardly to end exploiter , the level of scrutiny jumps tenfold , ” Sorensen said . “ Accuracy is the only thing that matter , because companies are disquieted about AI deceptive customers , and everyone has try on having ChatGPT or Claude hallucinate . A few bad response and a company will immediately suffer trustingness in your scheme . So that ’s what we handle about . ”
Accuracy
This focus on providing accurate responses about expert certification , with minimal delusion , play up how Kapa.ai is a unlike kind of LLM animal — it is make for a much narrow economic consumption case .
“ optimise a system for accuracy naturally comes with trade - offs , as it means we have to design the system to be less creative than what other LLM systems can afford to be , ” Sorensen said . “ This is to ensure the answers are only generated from the universe of subject they provide . ”
Then there is the thorny military issue of data point privacy — one of themajor deterrents for enterprisesthat maywantto espouse generative AI but are leery about exposing sensitive data to third - party organization . As such , Kapa.aiincludesPII ( personally identifiable information ) information - detection and masking , which extend some direction toward secure private information is neither hive away nor shared .
This includes real - time PII scanning : When a content is received by Kapa.ai , it ’s scanned for PII data , and if any personal data is detected , then the message is refuse and not salt away . Users can also configure Kapa.ai so that any PII information observe in a document will be anonymized .
Businesses can , of course , assemble something akin to Kapa.ai themselves using third - party tools such asAzure ’s OpenAI serviceorDeepset ’s Haystack . But it ’s a time - consuming and resource - intensive endeavor , especially when you’re able to just tap Kapa ’s website widget , deploy its bot for Slack or Zendesk , or use its API that leave troupe to customize things a small with their own interfaces .
“ Most of the people we work with do n’t want to do all the engineering work , or do n’t necessarily have the AI imagination on their teams to do so , ” Sorensen said . “ They want an exact and dependable AI locomotive engine that they can trust enough to reveal straight to client , and which has already been optimize for their use of goods and services case of answering technical product questions . ”
In terms of pricing , Kapa.ai says it uses a SaaS subscription model , offer up tiered pricing base on the complexness of the deployment and usage — though it does n’t publish these prices .
The company has a remote team of nine spread across the globe in two main hubs in Copenhagen , where Sorensen is based , and San Francisco .
Aside from lead backer Initialized Capital , Kapa.ai ’s source round construe participation from Y Combinator and a slew of angel investors , including Docker founder Solomon Hykes , Stanford professor and AI researcherDouwe Kiela , and Replit founder Amjad Masad .
