Funding for artificial intelligence increased last year, despite a general downturn in startup investments. From 2022 to 2023, funding for generative AI ventures alone than quadrupled, reaching $25.2 billion by the end of December.
Therefore, it should come as no surprise that AI startups topped the Winter 2024 Demo Day at Y Combinator.
As per YC’s official startup directory, the Winter 2024 cohort of Y Combinator has 86 AI businesses, which is almost twice as many as the Winter 2023 batch and nearly triple the amount of Winter 2021 startups. Whatever you think of AI, it’s undeniably the technology of the future.
We selected some of the most intriguing AI businesses from the most recent Y Combinator cohort, which is the cohort giving a presentation this week at Demo Day, just as we did the year before. Each was chosen for various reasons. However, based on their technology, target market, or founders’ backgrounds, they were unique from the competition.
The federal contracting process, according to Hazel August Chen (ex-Palantir) and Elton Lossner (ex-Boston Consulting Group), is utterly dysfunctional.
Contracts with hundreds of pages of conflicting regulations can be found on thousands of different websites. (It is believed that the U.S. federal government alone signs approximately 11 million contracts annually.) It may require entire corporate divisions to respond to these bids, with assistance from outside consultants and legal companies.
Hazel: The use of AI to automate the federal contracting process, including discovery, drafting, and compliance, is Chen and Lossner’s answer. The two, who were college friends, refer to it as Hazel.
With Hazel, users may match themselves with a possible contract, produce a draft response based on the details of their organization and the request for proposal, make a to-do list, and have compliance checks performed automatically.
I have some doubts about the accuracy of Hazel’s generated responses and checks, considering AI’s propensity for hallucinations. However, if they’re even close, they might save a ton of time and effort, giving smaller businesses a chance to compete for the government contracts worth hundreds of billions of dollars that are awarded every year.
Andy AI: Andy AI Home health aides handle a lot of documentation. Tiantian Zha is well-versed in this; she was involved in moonshots ranging from lowering mosquito-borne diseases to personalized medicine while she was employed at Verily, the life sciences branch of Alphabet, the parent company of Google.
Zha discovered during her employment that one of the biggest time wasters for at-home nurses was documentation. This is a common problem; a study found that nurses spend more than one-third of their time documenting, which reduces the amount of time they can spend caring for patients and increases the risk of burnout.
Zha and Max Akhterov, a former Apple staff engineer, co-founded Andy AI to lessen the paperwork burden for nurses. In essence, Andy is an AI-powered scribe who creates electronic health records and records and transcribing spoken facts from patient visits.
Bias is a possibility with any AI-powered transcription technology, meaning that certain nurses and patients may have difficulty using it due to word choice and accents. Furthermore, Andy isn’t exactly the first product of its ilk to hit the market; competitors include DeepScribe, Heidi Health, Nabla, and Amazon’s AWS HealthScribe.
However, if healthcare increasingly takes place at home, there will likely be a greater need for apps like Andy AI.
Precip: If you’ve used weather apps like this reporter, you’ve probably been caught in a downpour after naively trusting forecasts of sunny, clear sky.
However, things don’t have to be this way.
That is, at least, the theory behind Precip, a weather forecasting app driven by AI. The concept came to Jesse Vollmar when he started FarmLogs, a firm that offered crop management software. To make Precip a reality, he collaborated with Michael Asher, the former lead data scientist at FarmLogs, and Sam Pierce Lolla.
Precip provides analytics on precipitation, such as calculating the total amount of precipitation that has fallen in a specific location over the previous few hours to days. According to Vollmar, Precip can forecast circumstances up to seven days in advance and produce “high-precision” measurements for every area in the United States down to the kilometer (or two).
What use do precipitation metrics and notifications serve, then? According to Vollmar, utilities may use them to predict service interruptions, construction companies can use them to schedule crews, and farmers can use them to measure crop development. According to Vollmar, one transportation client analyzes Precip every day to prevent hazardous driving conditions.
Apps for predicting the weather are abundant, of course. However, if Precip’s AI is actually worth its salt, it should be able to improve forecast accuracy.
Maia: As Maia Claire Wiley was pursuing her MBA at Wharton, she started a couples coaching program. Her investigation into a more technologically advanced method to therapy and relationships followed the event, and the result was Maia.
Wiley and Ralph Ma, a former Google research scientist, co-founded Maia, a platform that uses AI to guide couples in forging stronger bonds. Couples message one other in a group chat on Maia’s applications for iOS and Android, answering daily questions like what obstacles they see ahead, past hurts, and lists of things they are grateful for.
Maia intends to monetize its offerings by charging for premium features like limitless chatting and programs created by therapists. (Maia now caps texts sent between couples; it’s a terribly arbitrary restriction, but that’s life.)
Wiley and Ma, who both hail from divorced families, claim that in order to create the Maia experience, they collaborated with a relationship expert. But the two things that are on my mind are (1) how reliable is Maia’s relationship science, and (2) is it able to differentiate itself in the incredibly competitive market for couples’ apps? We’ll have to wait and find out.
Datacurve: The generative AI models that power ChatGPT and other similar apps are trained on massive datasets that combine public and private material from the internet, such as social media posts, ebooks, and individual blogs. However, some of this data has ethical and legal issues in addition to other flaws.
For Serena Ge and Charley Lee, the issue is a glaring absence of data curation.
Datacurve, which offers “expert-quality” data for generative AI model training, was co-founded by Ge and Lee. It’s specifically code data, which Ge and Lee claim is particularly difficult to collect due to license restrictions and the skill required to classify it for AI training.
Datacurve provides training datasets for sale through a gamified annotation platform that compensates engineers for solving coding challenges. According to Ge and Lee, those datasets can be utilized to train models for debugging, UI design, code optimization, and more.
The concept is intriguing. However, Datacurve’s success will rely on how carefully chosen its datasets are and whether or not it can attract enough developers to keep enhancing and expanding them.
