AI Agents: plain-language access to clinical data
AI agents in Patient Journey Intelligence let clinicians, researchers, and care coordinators query patient information, build cohorts, and analyze outcomes by describing what they want in plain English. No SQL, no database schema, and no IT ticket.
The agents run on your standardized OMOP patient journeys. They read medical terminology, clinical workflows, and temporal relationships, and translate a question into a precise database query and a result in seconds.
Why clinical data access stalls
Four patterns come up in health systems that have already invested in EHRs and data warehouses:
Clinicians need answers, not SQL
A cardiologist wants to know which heart failure patients have not had an ejection fraction measured in the past year. Getting that answer through an IT ticket takes weeks, and getting it directly takes database skills that a clinical schedule leaves no room to develop.
Cohort identification takes days
Identifying trial-eligible patients by reviewing charts, exporting to spreadsheets, and applying inclusion and exclusion criteria by hand stretches a minutes-long question into days, which delays recruitment and caps how many studies a team can support.
Care teams cannot see the whole record
A patient's longitudinal journey, every diagnosis, medication change, lab result, and hospitalization, sits across dozens of EHR screens, external records, and legacy systems. Assembling that timeline by hand is slow and error-prone, and it is what disease progression, treatment response, and care gap review depend on.
Analysts become the bottleneck
When every clinical question, cohort request, and outcome analysis needs a custom query, results validation, and a report, urgent questions queue behind routine ones and a small change restarts the process.
How the agents remove those barriers
AI agents give clinicians and researchers self-service access to clinical data through conversational interfaces and guided workflows. Each one holds the clinical language and the database structure that answering a question requires, and is available whenever someone has a question to ask.
Plain-language clinical queries
Agents read medical terminology, clinical abbreviations, and healthcare workflows, so a question can be asked the way it would be asked of a colleague:
Show me diabetic patients with recent A1C greater than 9
The agent reads 'diabetic' as diabetes diagnosis codes, 'recent' as the interval clinical guidelines use (typically 3 to 6 months), and 'A1C > 9' as a specific lab measurement, then translates the question into OMOP queries across CONDITION_OCCURRENCE and MEASUREMENT.
What medications is patient MRN-12345 currently taking?
The agent retrieves active medications from DRUG_EXPOSURE, filters for current prescriptions on start and end dates, and presents drug names, doses, frequencies, and prescribing providers, with no manual EHR navigation.
Find patients eligible for colon cancer screening who are overdue
The agent applies age-based screening guidelines (typically 45 to 75 years), checks for a previous colonoscopy within the recommended interval (10 years at average risk), excludes patients with colorectal cancer history, and returns a cohort ready for outreach, from one conversational request.
Compare readmission rates between surgical and medical heart failure patients in the last year
The agent segments heart failure patients by treatment type, calculates 30-day readmission rates for each group, performs the statistical comparison, and generates visualizations, in place of hours of manual SQL and spreadsheet work.
Behind each of these questions the agent executes multi-table joins, applies clinical logic, and filters on standardized terminologies. None of that surfaces to the person asking.
Six specialized agents for common clinical workflows
Patient Journey Intelligence provides six specialized agents, each optimized for one clinical workflow and equipped with the tools that workflow needs. A single general-purpose assistant would carry the same tools for every job and fit none of them well.
Patient Copilot
Conversational assistant for querying individual patient data, exploring medical histories, and answering clinical questions in real time at the point of care.
Primary Users: Clinicians, nurses, care coordinators
Use Cases:
- Review lab result trends for a patient over a given period
- Check when a patient last had a screening or procedure
- List recent encounters, such as ER visits in the last six months
Learn more →
Patient Journey Viewer
Interactive timeline that assembles a complete longitudinal patient history, including diagnoses, medications, procedures, labs, and visits, into a narrative of disease progression and treatment response.
Primary Users: Care teams, utilization reviewers, case managers
Use Cases:
- Visualize multi-year patient timelines with all clinical events
- Filter by condition, medication class, or event type
- Identify care gaps and medication adherence patterns
Learn more →
Cohort Builder
No-code interface for defining patient populations with inclusion and exclusion criteria, temporal logic, and clinical concepts, so researchers and quality teams build complex cohorts without SQL.
Primary Users: Research coordinators, population health analysts, quality improvement teams
Use Cases:
- Identify trial-eligible patients for clinical studies
- Build quality measure cohorts (diabetes HbA1c control, HTN BP targets)
- Segment populations for targeted outreach campaigns
Learn more →
Agentic Search
Search a clinical document corpus before ingesting it. Query a PostgreSQL or Elasticsearch corpus in plain language, review matched patients with supporting evidence, and preview source documents before any data enters the platform.
Primary Users: Research coordinators, data engineers, registry teams
Use Cases:
- Confirm which patients are worth ingesting before running the full pipeline
- Preview matched documents from an external corpus before creating a dataset
- Search PostgreSQL or Elasticsearch corpora for clinical findings
Learn more →
Clinical Trial Matcher
Turns a trial protocol into patient matching logic. Parses ClinicalTrials.gov or pasted eligibility text into structured inclusion and exclusion criteria, runs the computable ones against OMOP data and clinical notes, and saves the matched patients as a reusable cohort.
Primary Users: Research coordinators, principal investigators, feasibility teams
Use Cases:
- Prescreen the local population against a protocol before screening visits
- Answer a sponsor's feasibility question from the full record, not diagnosis codes
- Test how a single criterion changes the eligible population
Learn more →
Care Gap Analyzer
Finds patients missing screenings, chronic disease monitoring, and guideline-recommended therapy. An auditable pipeline evaluates structured OMOP data and clinical notes against ACC/AHA, ADA, USPSTF, and HEDIS guidelines, with provenance on every finding.
Primary Users: Quality and population health teams, care management, payers
Use Cases:
- Find patients in a panel overdue for screening or monitoring
- Run HEDIS and CMS Star Ratings gap closure without a manual chart chase
- Track closure rates across successive runs on the same cohort
Learn more →
The six agents run on the same OMOP patient journeys, with Agentic Search reaching a step earlier to the corpus that has not been ingested yet. Each presents the interface its workflow needs and leaves out the controls it does not.
How an agent turns a question into an answer
Knowing how an agent gets from a question to a result is what lets a user trust the answer and ask a better question next time.
The Agent Query Workflow
A question is asked in plain language
A clinician or researcher types a question using medical terminology, clinical abbreviations, or plain English. No technical syntax is required.
The agent parses intent and clinical concepts
Language models trained on medical text identify the clinical intent ('find patients', 'show trends', 'compare outcomes') and extract the medical concepts mentioned: conditions, medications, procedures, and labs.
Concepts map to standardized terminologies
Clinical concepts are mapped to OMOP standard vocabularies: SNOMED CT for conditions, RxNorm for medications, LOINC for labs. The same query then works across institutions regardless of local coding practice.
OMOP queries are generated and executed
The agent constructs SQL against your OMOP CDM database, handling joins across PERSON, CONDITION_OCCURRENCE, DRUG_EXPOSURE, MEASUREMENT, and other tables, and applying the temporal filters and logic the question implies.
Results are formatted for clinical users
Raw database results are translated back into clinical language, formatted as tables or visualizations, and presented with the context needed to interpret them and act.
The workflow completes in seconds, which is what makes self-service clinical analytics usable during a clinic session.
Built on Model Context Protocol (MCP)
The agents are built on Model Context Protocol (MCP), an open standard for AI agent interoperability. MCP is what lets an agent invoke platform tools, reach external systems, and compose multi-step workflows without custom integration code.
What MCP enables for AI agents
Model Context Protocol lets agents discover and use available tools, whether they are built into Patient Journey Intelligence or provided by an external system. Agents can therefore:
- Invoke platform capabilities: query OMOP cohorts, compute clinical measures, export datasets, and generate reports through standardized MCP tools
- Integrate external systems: any MCP-compatible service, including scheduling systems, notification platforms, and EHR write-back APIs, becomes available to agents with no custom integration code
- Chain multi-step workflows: compose sequences such as "find eligible patients, check calendar availability, send recruitment letters, track responses" by orchestrating several MCP tools
- Add custom extensions: build MCP tools that expose institutional logic such as IRB approval workflows, data access controls, or local coding systems, and agents learn to use them
Example: a research coordinator asks an agent to "identify eligible patients for the diabetes trial and draft recruitment letters." The agent uses MCP to query the OMOP cohort, retrieve patient contact information, generate personalized letters with a document template tool, and queue them in an email system, with no code written by the coordinator.
Standardizing on MCP is what keeps the agents open to your existing systems and to tools that do not exist yet, without vendor lock-in.
What self-service access changes
Conversational access to patient data removes the queue between a clinical question and its answer.
Minutes instead of days
Cohort identification that took days of chart review or time in an IT backlog completes in minutes. Research coordinators answer feasibility questions during the call, which moves trial recruitment and grant applications forward.
Answers at the point of care
Physicians query patient histories, medication interactions, and lab trends during a clinic visit, without leaving the exam room or waiting for chart review. Longitudinal data reaches the conversation while it can still change it.
Analysts work on strategy
Self-service access clears the backlog of routine data requests, so analysts move from repetitive SQL to designing quality measures, validating outcomes, and answering novel research questions.
Questions that were too expensive to ask
When a clinical question costs days of manual work, most questions are never asked. Plain-language access makes exploratory analysis cheap enough to pursue a hunch, which is where patterns and opportunities are found.
Standardized results, zero SQL errors
Agents apply consistent logic, terminologies, and temporal filters across every query, which removes the inconsistencies that appear when each analyst writes their own version of 'diabetic patients' or 'recent lab results.'
Lower training burden for new staff
New research coordinators, quality analysts, and care managers become productive without SQL training, schema documentation, or months of institutional knowledge transfer. Describing what they need is enough.
Getting started with AI agents
The agents are available as soon as your OMOP patient journeys are established. They discover your data schema and begin answering questions, with no separate configuration, model training, or integration work.
Start with Patient Copilot
Query individual patient data in plain language for clinical decision support and care coordination.
Try Patient Copilot →
Explore patient journeys
Visualize longitudinal patient timelines showing complete medical histories and disease progression.
View Patient Journeys →
Build your first cohort
Define patient populations for research, quality measures, or targeted interventions without writing code.
Build a Cohort →
Search before you ingest
Query an external PostgreSQL or Elasticsearch document corpus in plain language and review matched patients before any data enters the platform.
Try Agentic Search →
Screen for a trial
Turn a ClinicalTrials.gov protocol or pasted eligibility text into patient matching logic, and save the matched patients as a cohort.
Match a Trial →
Find care gaps
Identify patients missing screenings, monitoring, or guideline-recommended therapy, with the evidence behind every finding.
Analyze Care Gaps →
Six: Patient Copilot for questions about an individual patient, Patient Journey Viewer for longitudinal timelines, Cohort Builder for no-code population definitions, Agentic Search for querying an external clinical document corpus before ingestion, Clinical Trial Matcher for screening patients against trial eligibility criteria, and Care Gap Analyzer for finding patients missing guideline-recommended care. All run on your OMOP patient journeys except Agentic Search, which runs against a registered corpus that has not been ingested yet.
No. Questions are asked in plain language, and the agent parses the clinical intent, maps concepts to OMOP standard vocabularies (SNOMED CT, RxNorm, LOINC), constructs the SQL against your OMOP CDM database, and formats the result in clinical language.
Clinical concepts are mapped to OMOP standard vocabularies before a query runs, so a query written at one institution works at another regardless of local codes. This is also what keeps two analysts from producing different definitions of the same population.
MCP lets agents discover and invoke tools, both the platform's own and any MCP-compatible external service such as a scheduling system, notification platform, or EHR write-back API. It also lets you expose institutional logic, such as IRB approval workflows or local coding systems, as custom tools that agents learn to use.
None beyond having your OMOP patient journeys established. The agents discover the data schema and start answering questions, with no separate configuration, model training, or integration work.