Introduction

Earlier this year, I set about creating kit to introduce young techy folk at a Health Tech hackathon to using InterSystems IRIS for health, particularly focusing on using FHIR and vector search.

I wanted to publish this to the developer community because the tutorials included in the kit make a great introduction to using FHIR and to building a basic RAG system in IRIS.

What is Unstructured Data?
Unstructured data refers to information lacking a predefined data model or organization. In contrast to structured data found in databases with clear structures (e.g., tables and fields), unstructured data lacks a fixed schema. This type of data includes text, images, videos, audio files, social media posts, emails, and more.

Why Are Insights from Unstructured Data Important?
According to an IDC (International Data Corporation) report, 80% of worldwide data is projected to be unstructured by 2025, posing a significant concern for 95% of businesses.

In the previous article, we saw in detail about Connectors, that let user upload their file and get it converted into embeddings and store it to IRIS DB. In this article, we'll explore different retrieval options that IRIS AI Studio offers - Semantic Search, Chat, Recommender and Similarity. 

New Updates  ⛴️ 

• Added installation through Docker. Run `./build.sh` after cloning to get the application & IRIS instance running in your local
• Connect via InterSystems Extension in vsCode - Thanks to @Evgeny.

Keywords:  Vibe coding, Windsurf, IRIS, TIE

Why not?   "Vibe coding" is never about the vibe!

Has anyone not been trying "vibe coding" so far?

Even merely 3 years ago, if anyone asked

• "Could I do IRIS implementation for NHS TIE in English or Spanish or just Chinese ?", or
• "Can I just instruct TIE in English to build itself an e2e route, to pick up a PDF report then turn into ORU/MDM message and submit into the PAS ?", or
• "Could we query IRIS database in English only, and build up dashboard or ad hoc report of my own by English instructions?

Keywords:  IRIS, Agents, Agentic AI, Smart Apps

Motive?

Transformer based LLMs appear to be a pretty good "universal logical–symbolic abstractor".  They started to bridge up the previous abyss among human languages and machine languages, which in essence are all logic symbols that could be mapped into the same vector space. 

Objective?

Wondering for 3 years we might be able to just use English (etc human natural languages) to do IRIS implementations as well, one day.

Ever since I started using IRIS, I have wondered if we could create agents on IRIS. It seemed obvious: we have an Interoperability GUI that can trace messages, we have an underlying object database that can store SQL, Vectors and even Base64 images. We currently have a Python SDK that allows us to interface with the platform using Python, but not particularly optimized for developing agentic workflows. This was my attempt to create a Python SDK that can leverage several parts of IRIS to support development of agentic systems.

10:47 AM — Jose Garcia's creatinine test results arrive at the hospital FHIR server. 2.1 mg/dL — a 35% increase from last month.

What happens next?

• Most systems: ❌ The result sits in a queue until a clinician reviews it manually — hours or days later.
• This system: 👍 An AI agent evaluates the trend, consults clinical guidelines, and generates evidence-based recommendations — in seconds, automatically.

No chatbot. No manual prompts. No black-box reasoning.

This is event-driven clinical decision support with full explainability:

✅ Triggered automatically by FHIR events ✅ Multi-agent reasoning (context, guidelines, recommendations) ✅ Complete audit trail in SQL (every decision, every evidence source) ✅ FHIR-native outputs (DiagnosticReport published to server)

Built with:

• InterSystems IRIS for Health — Orchestration, FHIR, persistence, vector search
• CrewAI — Multi-agent framework for structured reasoning

You'll learn: 🖋️ How to orchestrate agentic AI workflows within production-grade interoperability systems — and why explainability matters more than accuracy alone.

How I Vibecoded a Backend (and Frontend) on InterSystems IRIS

I wanted to try vibecoding a real backend + frontend setup on InterSystems IRIS, ideally using something realistic rather than a toy example. The goal was simple: take an existing, well-known persistent package in IRIS and quickly build a usable UI and API around it — letting AI handle as much of the boilerplate as possible. Here is the result of the experiments.

Vector search is a retrieval method that converts text, images, audio, and other data into numeric vectors using an AI model, and then searches for items that are semantically close. It enables “semantic similarity search” from free text, which is difficult with keyword search alone.

However, in real use, I encountered cases where results that are “close in meaning” but logically the opposite appeared near the top of the search results.

This is a serious issue in situations where affirmation vs. negation matters. If the system returns the wrong answer, the impact can be significant, so we cannot ignore this problem.

This article does not propose a new algorithm. I wrote it to share a practical way I found useful when semantic search fails due to negation.

Some concepts make perfect sense on paper, whereas others require you to get your hands dirty. Take driving, for example. You can memorize every component of the engine mechanics, but that does not mean you can actually drive.

You cannot truly grasp it until you are in the driver's seat, physically feeling the friction point of the clutch and the vibration of the road beneath. While some computing concepts are intuitive, Intelligent Agents are different. To understand them, you have to get in the driver's seat.

Introduction

As AI-driven automation becomes an essential part of modern information systems, integrating AI capabilities into existing platforms should be seamless and efficient. The IRIS Agent project showcases how generative AI can work effortlessly with InterSystems IRIS, leveraging its powerful interoperability framework—without the need to learn Python or build separate AI workflows from scratch.
This project explores how ChatGPT and Anthropic Claude, two of the most advanced AI models, can interact with IRIS using Custom GPTs and the Model Context Protocol (MCP).

How to set up RAG for OpenAI agents using IRIS Vector DB in Python

In this article, I’ll walk you through an example of using InterSystems IRIS Vector DB to store embeddings and integrate them with an OpenAI agent.

To demonstrate this, we’ll create an OpenAI agent with knowledge of InterSystems technology. We’ll achieve this by storing embeddings of some InterSystems documentation in IRIS and then using IRIS vector search to retrieve relevant content—enabling a Retrieval-Augmented Generation (RAG) workflow.

Note: Section 1 details how process text into embeddings.

Vibe the Module, Not the Data

While working with the FHIR to OMOP Service, I've seen good FHIR synthetic data being created using commercial LLM's etc, custom tailored for ConditionOnset with the typical amazement on return, but witnessed some questionable trust first hand on a call.  This approach also falls short generating gigantic payloads so I can go back to my interests on the backend and ensure smooth data transition.

Vanna.AI - Personalized AI InterSystems OMOP Agent

Along this OMOP Journey, from the OHDSI book to Achilles, you can begin to understand the power of the OMOP Common Data Model when you see the mix of well written R and SQL deriving results for large scale analytics that are shareable across organizations. I however do not have a third normal form brain and about a month ago on the Journey we employed Databricks Genie to generate sql for us utilizing InterSystems OMOP and Python interoperability.

In the previous article, we saw how to build a customer service AI agent with smolagents and InterSystems IRIS, combining SQL, RAG with vector search, and interoperability.

In that case, we used cloud models (OpenAI) for the LLM and embeddings.

This time, we’ll take it one step further: running the same agent, but with local models thanks to Ollama.

Plug-N-Play on Pattern Match WorkBench

Article to announce pre-built pattern expressions are available from demo application.

AI deducing patterns require ten and more sample values to get warmed up.

The entry of a single value for a pattern has therefore been repurposed for retrieving pre-built patterns.

Example: Email address

Paste an sample value for example an email address in description and press "Pattern from Description".

The sample is tested against available built-in patterns and any matching patterns and descriptions are displayed.

Patterns can also be retrieved by Keyword.

If you’ve ever watched a true artisan—whether a potter turning mud into a masterpiece or a luthier bringing raw wood to life as a marvelous guitar—you know that magic isn’t in the materials, but in care, craft, and process. I know this firsthand: my handmade electric guitar is a daily inspiration, but I’ll admit—creating something like that is a talent I don’t have.

Yet, in the digital world, I often see people hoping for “magic” from generative AI by typing vague, context-free prompts like “build an app.” The results are usually frustratingly shallow—no artistry, no finesse.

Thirteen years ago, I attained dual undergraduate degrees in electrical engineering and math, then promptly started full-time at InterSystems using neither. One of my most memorable and stomach-churning academic experiences was in Stats II. On an exam, I was solving a moderately difficult confidence interval problem. I was running out of time, so (being an engineer) I wrote out the definite integral on the exam paper, punched it into my graphing calculator, wrote an arrow with “calculator” over it, then wrote the result.

Audience

Those curious in exploring new GenerativeAI usecases.

Shares thoughts and rationale when training generative AI for pattern matching.

Challenge 1 - Simple but no simpler

A developer aspires to conceive an elegant solution to requirements.
Pattern matches ( like regular expressions ) can be solved for in many ways. Which one is the better code solution?
Can an AI postulate an elegant pattern match solution for a range of simple-to-complex data samples?

☤ Care 🩺 Compass 🧭 - Proof-of-Concept - Demo Games Contest Entry

Introducing Care Compass: AI-Powered Case Prioritization for Human Services

In today’s healthcare and social services landscape, caseworkers face overwhelming challenges. High caseloads, fragmented systems, and disconnected data often lead to missed opportunities to intervene early and effectively. This results in worker burnout and preventable emergency room visits, which are both costly and avoidable.

Care Compass was created to change that.

Thank you community for translating an earlier article into Portuguese.
Am returning the favor with a new release of Pattern Match Workbench demo app.

Added support for Portuguese.

The labels, buttons, feedback messages and help-text for user interface are updated.

Pattern Descriptions can be requested for the new language.

The single AI Model for transforming user prompt into Pattern match code was fully retrained.

Values to Pattern Code Model also retrained

The separate AI model for generating Pattern match code from a sample list of values has been retrained.

INTRO

Barricade is a tool developed by ICCA Ops to streamline and scale support for FHIR-to-OMOP transformations for InterSystems OMOP. Our clients will be using InterSystems OMOP to transform FHIR data to this OMOP structure. As a managed service, our job is to troubleshoot any issues that come with the transformation process. Barricade is the ideal tool to aid us in this process for a variety of reasons. First, effective support demands knowledge across FHIR standards, the OHDSI OMOP model, and InterSystems-specific operational workflows—all highly specialized areas.

Audience

Those curious in exploring new GenerativeAI usecases.

Developers and analysts looking for a quick way to tame the Pattern Match operator.

In both ObjectScript and SQL this has a quite visually dense format.

 if booking?2A1"-"1(1"CARD",1"RAD")1.5N

SELECT DISTINCT PatientRef
FROM APPOINTMENT.BOOKING
WHERE Active='Y' AND
LocationCode %PATTERN '2A1"-"1(1"CARD",1"RAD")1.5N'

Challenge

Use generative AI to assist create and modify pattern match code from different natural language input.

Artificial Intelligence (AI) is getting a lot of attention lately because it can change many areas of our lives. Better computer power and more data have helped AI do amazing things, like improving medical tests and making self-driving cars. AI can also help businesses make better decisions and work more efficiently, which is why it's becoming more popular and widely used. How can one integrate the OpenAI API calls into an existing IRIS Interoperability application?

Hi developers!

This will be a very short article as in April 2025 with Lovable and other Prompt-to-UI tools it becomes possible to build the frontend with prompting. Even to the folks like me who is not familiar with modern UI techics at all.

Well, I know at least the words javascript, typescript and ReactJS, so in this very short article we will be building the ReactJS UI to InterSystems FHIR server with Lovable.ai.

Let's go!

Hi, colleagues!

As you can see the new topic of the programming contest - AI Agents. 

The topic was over-hyped recently on the Internet and has different meanings. You might get curious about what we mean by AI agents in regard to the InterSystems programming contest.