Featured
Xyonix recently developed a cutting-edge virtual concierge system for deployment across thousands of hotels worldwide. This AI-driven system features a sophisticated knowledgebase for hotel-specific inquiries and an LLM-based dialog system for dynamic guest interaction. The system also includes advanced features to avoid bot hallucinations and maintain creative control over responses, offering broad guidance in general queries while relying on a curated knowledgebase for specific information. Additionally, it intelligently classifies guest questions, seamlessly directing complex issues to human staff. [Read more]
For an extremely rapidly growing startup, we built an AI system that transforms lengthy, often instructional videos into concise, optimized segments. The system cleanses transcripts, then leverages LLM models we fine tuned using training data meticulously annotated by a combination of other LLMs and our human team at Xyonix. It's adept at extracting key topics, generating summaries, easing navigation, and producing short, impactful videos in various formats. Already in public use and benefiting millions daily, this system is a game-changer for educational and instructional content, making complex information more accessible and engaging.
We worked closely with a company that helps colleges and universities better advise their student populations in a variety of areas including career choices, adjusting to college life, finding motivation and more. We built a virtual concierge like application that transcends traditional question/answering engaging in highly sophisticated and helpful conversation. We worked closely with the company CEO and product teams to think through a variety of advanced applications, build demonstrable prototypes, and facilitate easy deployment into production. [Read more]
In collaboration with a rapidly growing startup, we've crafted an AI-driven solution for helping users rapidly create high-quality video content by automatically analyzing user generated movie scripts and identifying optimal video assets for inclusion in short videos. Our role extended beyond data science; we engineered and now host a fully scaled system. This platform excels in generating scene text and keywords, leveraging semantic search within a bespoke imagery index, and creating video metadata from video frames. Designed to facilitate the production of engaging videos and scripts for social media and advertising, this service is robust, fully managed, and caters to millions of users daily.
We were asked by Columbia University to help build a map of vaccine hesitancy to assist public health officials in increasing vaccination rates. We built a corpus of millions of social media messages from platforms like Twitter, Reddit and Youtube comments. We are now analyzing this data by manually annotating thousands of training examples using a multi-parent taxonomy and iteratively (active learning driven) training a multi-label machine learning and NLP powered parser. If we are successful, we hope to save lives by convincing the vaccine hesitant to protect themselves and their communities. [Read more]
We helped a company developing a virtual AI tool for improving student progression understand how to facilitate interactions between students, parents, and teachers to optimize teaching outcomes. We proposed a system with the ability to analyze factors affecting student outcomes and use that information to instruct teachers and parents on how to best observe and assist their students.ns, and teacher actions. It could then use this knowledge to instruct teachers and parents on how to observe student progress and provide them with specific feedback on observational ability, engagement, and teaching plan quality.
We helped a company build a powerful virtual intake specialist that automatically communicates with patients to route them to the best care for their condition. Our AI system listens to patients and asks them just the right follow up questions, much like a traditional intake nurse might, to help schedule the best care option. [Read more]
Xyonix teamed up with an early-stage startup to launch an AI iPhone app based assistant to help Med Spa nurse practitioners. The AI technology adeptly categorized facial features, aiding in the assessment of treatments and spotting issues like temporal visual changes, ineffective product batches or injector performance. Integrated with iOS for precise patient driven photo alignment, the system enabled consistent treatment tracking. We played a pivotal role in technology development and strategic product planning; our involvement was key in fusing medical accuracy with technological innovation in product development. [Read more]
For a fortune 500 company, our Xyonix built models are in production today constantly analyzing thousands of recent in body surgery videos and reviews across many hundreds or thousands of surgery rooms around the world. Our models watch never ending streams of surgery videos, and read countless reviews. Our models power dashboards that help surgeons improve. [Read more]
We worked closely with a large industrial parts marketplace provider to help them better understand their current customers and partners. We developed AI models trained on their customer and partner site usage, profile and external attribute data to analyze engagement. We were able to successfully forecast which customers and partners were destined for high and low engagement. Perhaps more importantly, our models were able to explain why customers were likely to be high or low engagers. These deep, data driven insights, when used to power a living intelligence dashboard, are able to direct sales and marketing activities so they are more efficient and impactful.
We worked very closely with executives of a major dental insurance company to help better understand their customers. We used historic data to predict future customer behavior and purchase selections. Our models predict which customers are likely to lapse, and which are likely to remain loyal. Our models are also explainable, so it is easier to understand which attributes of a customer are indicative of loyalty or lapse tendencies. In addition to understanding high level customer characterizations, our explainable models describe why a specific customer is predicted to be a lapser or loyalist.
We worked closely with a company to empower consumers with readily accessible quality health information by building systems that read large volumes of unstructured medical texts using deep NLP relation parsing techniques, and automatically authored millions of web pages. Our work resulted in site traffic growth of 10,000% averaging 6% week over week growth. The site attracted primarily physicians, physician specialists, and other researchers due in large part to the quality of the automatically generated text.
We helped our client build out automated tools for early motor development risk and autism spectrum screening. Automatically and accurately extracting an infant’s pose and understanding these movements over time can be essential in preventing a lifetime of problems.
We built a system that helped our client help popular artists like Metallica and Beyonce communicate with their massive fan base by automatically interpreting millions of fan text messages sent to artist’s personal phones. Our parser was capable of accurately recognizing over 70 detailed conversation statements.
We worked closely with executives and data scientists of a major automated sales lead generation system to define an AI path forward that enabled their system to not only understand conversations, but to also generate responses for human consumption. We were able to efficiently extend one conversational domain from the recognition of just a handful of response categories, to over 50, all with very high accuracy.
We built a system that helps artists like Metallica and Beyonce interpret the results of millions of fan text messages. During our analysis, we found that many fans were expressing depression related thoughts. We honed our AI models to more accurately differentiate between those discussing sad things and those more at risk of self harm. The results help get these fans professional help.
We helped a widely used patient population software management company predict the number of patients that will be arriving at a given unit of a hospital. Our deep learning based solution accurately forecasts the patient census at various hourly intervals up to 5 days into the future.
We trained an AI system to automatically detect cancerous lesions from a customer’s novel disposable endoscope. Our system can help physicians detect lesions including papillary carcinomas in procedures like a cytoscopy.
We enabled a smartphone based digital stethoscope with AI to assist general physicians and cardiologists in diagnosing anomalous heart conditions including heartbeat murmurs. Our client’s now AI powered stethoscope is used daily by many physicians to detect potential problems in patients. Demo video HERE.
Can we better an individual goalie’s chances of stopping a penalty kick with advanced AI? Many of us at Xyonix are huge soccer fans. With the recent excitement of the 2019 FIFA Women’s World Cup, and the success of the U.S. Women’s National Team, we got inspired to try and apply advanced AI techniques to soccer. The results helped us work out the kinks in our activity analysis platform.