Ontash

Artificial Intelligence @ Ontash

AI at Ontash Hero Image

Our engineers have been involved in AI research since the inception of the field. We have systems that use AI in production for over 8 years at major hospital systems delivering measurable impact on the lives of over a thousand patients while increasing hospital revenue.

Uncompensated Care Disability Services (UCDS) is a software platform that hospitals use to identify uninsured patients with AI assistance, and then manage the workflow needed to qualify them for SSI and SSDI benefits. This allows the hospital to receive payment for services (through Medicare or Medicaid) that would otherwise be written‑off. Patients could also receive cash benefits. For more on this usage of AI in Healthcare see: UCDS - A Case Study of AI Usage in Healthcare.

Artificial Intelligence (AI) is software designed to mimic human intelligence—enabling computers to learn, reason, solve problems, make decisions, and even create content, much like people do. What counts as “AI” is always shifting: technologies once considered AI, like basic speech recognition or chess‑playing programs, are now seen as routine, while today’s advanced systems—such as large language models (LLMs) and retrieval‑augmented models (RLMs)—may soon become so common they’re no longer labeled as AI. This field has advanced rapidly, mainly due to exponential increases in computing power (faster chips, widespread use of GPUs, more memory), which allow machines to process vast amounts of data and perform ever more complex tasks

AI is not a panacea for every problem, but rather a tool which will rapidly become an integral part of our lives and businesses. The key is to understand where AI can be used, how to use it and lastly, to use it cost effectively. Ontash can help you with:

  • Identifying AI Opportunities: Understanding your business processes and pain points to determine those processes that could be enhanced by AI
  • AI Powered Business Automation: Build AI software and/or Agents to replace or enhance your business processes
  • AI tooling and infrastructure: Building low level tools and computing infrastructure that will be used to build and run your AI Agents

Identifying AI Opportunities

It is critical to first understand your business processes and pain points to determine what specifically could be enhanced by AI. Once such opportunities are identified they need to be validated by estimating the effectiveness of the intervention, and the costs involved. In many situations a rapid prototype or proof of concept can be used.

Some examples of situations where AI could be used:

  • Predictive Modeling and Forecasting model and forecast sales, engagement, medical, financial and other data
  • Customer Outreach, Market Research, and Streamlined Engagement
  • Claims & Application Processing (e.g., medical insurance, social security disability)
  • Document Comprehension and Summarization (including for disability applications)
  • Validating Computer Code for conformance with Best Practices
  • Smart Chatbot trained on your internal documents, capable of answering questions and generating images based on user prompts via voice, sms and text interface
  • Face and Object Recognition (including license plate recognition, employee/visitor recognition, and manufacturing quality assurance)
  • License Plate Recognition System for building facilities and traffic automation

AI Powered Business Automation

Build AI software and/or Agents to replace or enhance your business processes

It is very important to understand that it can be deceptively easy to come up with an AI solution that appears to work. Putting the software into production and achieving the last 20% of accuracy can take a prohibitively long period of time.

In many cases it will not be wise to rely entirely on AI for a function but instead build a process where AI can make some decisions with human supervision. Alternatively if AI is the main decision maker in a process you have to plan appropriate safeguards for those cases where AI will inevitably make erroneous decisions.

AI Software performance also requires measuring effectiveness and monitoring over time, as the software and tools (LLM’s, and other AI Models) are continually upgraded.

AI Software has to be written in such a way, that it improves as rapidly as the tools on which it is built improves - which is very fast.

Ontash can guide you in the right direction.

AI Infrastructure and Tooling

AI Software has many dependencies which can be external in the cloud or on premise. These tools themselves can be open source, proprietary or internally developed. There are significant cost, performance and security implications in the choice of infrastructure and tooling used to build a particular software. For instance a proprietary AI model may not give you the desired accuracy and may cost too much to run. In healthcare applications one has to prioritize the safety of data, and one way of doing so, is to use an on-premises AI engine. Ontash can help you make these decisions and also implement appropriate tools and infrastructure including:

  • On-Premises AI Inference and Training Solutions for data security and cost optimization
  • AI Engines: Currently LLM (Large Language Models) are all the rage but in many situations other technologies are adequate:
    • Rule‑Based Systems
    • Traditional Machine Learning Models (e.g., SVMs, Decision Trees)
    • LLM (Large Language Model) with Fine‑Tuning
  • AI Engine Plug‑Ins: These are interfaces between AI engines and non‑AI software. These interfaces allow AI engines to retrieve and manipulate information on legacy systems using natural language. There are different standards for such plug‑ins including MCP (Module Context Protocol) Servers and LLM Tools (Check this term?). Ontash can implement these plug‑ins for your use cases.

UCDS - Case Study of AI Usage in Healthcare

Here is a real-world example of an application which uses AI: Uncompensated Care Disability Services (UCDS) is a software platform that hospitals use to identify uninsured patients with AI assistance, and then manage the workflow needed to qualify them for SSI and SSDI benefits. This allows the hospital to receive payment for services (through Medicare or Medicaid) that would otherwise be written‑off. Patients could also receive cash benefits.

The application has been in production for over eight years and incorporates the use of many different AI technologies.

Identification of disabled patients from patient data feed

UCDS AI software Inspects a data feed of patient information including demographics, and diagnostic codes to determine patients likely to meet disability criteria. The data is run through two AI engines (1) a simple rule engine; and (2) a Machine Learning (ML) model. he rules engine was created using input from doctors. The ML engine was trained using historical data.

Screening of patients via voice AI Agent

Likely candidates need to be contacted and screened before starting the disability application process. This process is now done by a voice AI agent which performs the screening before passing the candidate on to a case worker.

Recognition and Dispatching of Incoming Documents

The process of managing a large number of disability applications by using non‑AI software is very labor intensive. The large number of documents being received by mail include medical records, awards, denial notices and requests for more information. Previously this work was done manually. Now letters are opened and fed into a scanner where aLarge Language Model based AI program identifies the document, which claimant the document is associated with, then loads it to the appropriate place in the software and posts tasks for any follow‑up actions.

Summarization of medical records

An important part of the application process is collecting supporting medical evidence from the patient’s various doctors. Medical records are not concise, received in different formats and full of duplicated information. UCDS uses an LLM based agent to scan all the received documents and create a summary which references the underlying documents. This greatly reduces the time the case worker spends on preparing the application and more importantly makes it easier for workers at the Social Security Administration to understand it.

Quality Checks of a Case History

The disability application process is complex and requires a great deal of attention by the case worker to keep the case history complete. However data entry omissions are common for a variety of reasons and are often not detected until the case is seriously overdue for resolution. The solution was to use an LLM based AI agent to scan the case history every few days to look for problems, and where possible, resolve the problem automatically and alert the assigned case manager. This has reduced the number of errors and delays in processing cases.