Transformers

Transformers are the core logic of every channel. Written in TypeScript, they receive an incoming message, manipulate it however you need, and return the result to continue through the pipeline. Because transformers are plain TypeScript functions, you have the full npm ecosystem at your disposal — parse HL7v2 with node-hl7-client, build FHIR resources, convert CSV, reshape JSON, or call any third-party API.

Function Signature

Every transformer exports a single transform function:

export function transform(
  msg: IntuMessage,
  ctx: IntuContext
): IntuMessage | IntuMessage[] | null;

Parameter	Type	Description
msg	IntuMessage	The inbound message with rawData, content, headers, and metadata
ctx	IntuContext	Pipeline context — map variables, logger, channel config

Return Types

The return type controls how the pipeline handles the transformer output:

Return	Behavior	Use Case
IntuMessage	One message continues through the pipeline	1:1 transformation
IntuMessage[]	Each element continues independently	Fan-out / batch splitting
null	Message is silently dropped	Filtering / conditional routing

// 1:1 — transform a single message
export function transform(msg: IntuMessage, ctx: IntuContext): IntuMessage {
  msg.content = JSON.stringify({ wrapped: JSON.parse(msg.rawData) });
  return msg;
}

// 1:many — fan out a FHIR Bundle into individual resources
export function transform(msg: IntuMessage, ctx: IntuContext): IntuMessage[] {
  const bundle = JSON.parse(msg.rawData);
  return bundle.entry.map((entry: any) => ({
    ...msg,
    content: JSON.stringify(entry.resource),
  }));
}

// Filter — drop messages that don't match criteria
export function transform(msg: IntuMessage, ctx: IntuContext): IntuMessage | null {
  const hl7 = msg.rawData;
  if (!hl7.startsWith("MSH|")) return null;
  return msg;
}

Auto-Compilation and Hot Reload

TypeScript is compiled automatically when the engine starts. During development (npm run dev), the engine watches for file changes and recompiles on save — no manual build step required.

Tip

Edit a .ts file and save. The channel restarts within milliseconds and picks up your changes immediately.

Type Safety with intu.d.ts

Every scaffolded project includes src/types/intu.d.ts, which declares the IntuMessage and IntuContext interfaces. Your editor provides autocomplete, type checking, and inline documentation out of the box.

interface IntuMessage {
  rawData: string;
  content: string;
  headers: Record<string, string>;
  sourceType: string;
  channelId: string;
  messageId: string;
  timestamp: string;
}

interface IntuContext {
  logger: IntuLogger;
  globalMap: Map<string, any>;
  channelMap: Map<string, any>;
  responseMap: Map<string, any>;
  connectorMap: Map<string, any>;
  channelId: string;
  channelName: string;
}

Note

intu.d.ts is generated by intu init. Keep it in your repo so every team member and CI environment shares the same type definitions.

Using the Full npm Ecosystem

Transformers run in a standard Node.js environment. Install any npm package and import it directly:

npm install node-hl7-client fhir fast-csv fast-xml-parser

import { Hl7Message } from "node-hl7-client";

export function transform(msg: IntuMessage, ctx: IntuContext): IntuMessage {
  const hl7 = new Hl7Message(msg.rawData);
  const patientName = hl7.get("PID.5");
  const mrn = hl7.get("PID.3.1");

  msg.content = JSON.stringify({
    resourceType: "Patient",
    identifier: [{ system: "http://hospital.example/mrn", value: mrn }],
    name: [{ family: patientName.split("^")[0], given: [patientName.split("^")[1]] }],
  });

  return msg;
}

Examples

HL7v2 ADT to FHIR Patient

import { Hl7Message } from "node-hl7-client";

export function transform(msg: IntuMessage, ctx: IntuContext): IntuMessage {
  const hl7 = new Hl7Message(msg.rawData);

  const pid = {
    family: hl7.get("PID.5.1"),
    given: hl7.get("PID.5.2"),
    dob: hl7.get("PID.7"),
    gender: hl7.get("PID.8"),
    mrn: hl7.get("PID.3.1"),
    ssn: hl7.get("PID.19"),
    address: {
      line: hl7.get("PID.11.1"),
      city: hl7.get("PID.11.3"),
      state: hl7.get("PID.11.4"),
      postalCode: hl7.get("PID.11.5"),
    },
  };

  const patient = {
    resourceType: "Patient",
    identifier: [
      { system: "http://hospital.example/mrn", value: pid.mrn },
      { system: "http://hl7.org/fhir/sid/us-ssn", value: pid.ssn },
    ],
    name: [{ family: pid.family, given: [pid.given] }],
    birthDate: `${pid.dob.slice(0, 4)}-${pid.dob.slice(4, 6)}-${pid.dob.slice(6, 8)}`,
    gender: pid.gender === "M" ? "male" : pid.gender === "F" ? "female" : "unknown",
    address: [
      {
        line: [pid.address.line],
        city: pid.address.city,
        state: pid.address.state,
        postalCode: pid.address.postalCode,
      },
    ],
  };

  msg.content = JSON.stringify(patient);
  return msg;
}

CSV Lab Results to FHIR Observations

import { parse } from "fast-csv";

export async function transform(
  msg: IntuMessage,
  ctx: IntuContext
): Promise<IntuMessage[]> {
  const rows: any[] = [];

  await new Promise<void>((resolve, reject) => {
    const stream = parse({ headers: true }).on("data", (row) => rows.push(row)).on("end", resolve).on("error", reject);
    stream.write(msg.rawData);
    stream.end();
  });

  return rows.map((row) => ({
    ...msg,
    content: JSON.stringify({
      resourceType: "Observation",
      status: "final",
      code: {
        coding: [{ system: "http://loinc.org", code: row.loinc_code, display: row.test_name }],
      },
      subject: { reference: `Patient/${row.patient_id}` },
      valueQuantity: {
        value: parseFloat(row.value),
        unit: row.unit,
        system: "http://unitsofmeasure.org",
        code: row.unit,
      },
      effectiveDateTime: row.collected_date,
    }),
  }));
}

JSON Payload Reshape

export function transform(msg: IntuMessage, ctx: IntuContext): IntuMessage {
  const input = JSON.parse(msg.rawData);

  msg.content = JSON.stringify({
    id: input.patient_id,
    fullName: `${input.first_name} ${input.last_name}`,
    dob: input.date_of_birth,
    facility: input.sending_facility ?? "UNKNOWN",
    receivedAt: msg.timestamp,
  });

  return msg;
}

XML CDA to JSON

import { XMLParser } from "fast-xml-parser";

export function transform(msg: IntuMessage, ctx: IntuContext): IntuMessage {
  const parser = new XMLParser({ ignoreAttributes: false });
  const doc = parser.parse(msg.rawData);

  const patient = doc.ClinicalDocument?.recordTarget?.patientRole?.patient;

  msg.content = JSON.stringify({
    resourceType: "Patient",
    name: [
      {
        family: patient?.name?.family,
        given: [patient?.name?.given],
      },
    ],
    gender: patient?.administrativeGenderCode?.["@_code"] === "M" ? "male" : "female",
    birthDate: patient?.birthTime?.["@_value"],
  });

  return msg;
}

Map Variables

Map variables let you share state across pipeline stages. They are available on ctx in every transformer:

Map	Scope	Lifetime	Use Case
globalMap	All channels	Engine lifetime	Lookup tables, shared config
channelMap	Single channel	Channel lifetime	Sequence counters, caches
responseMap	Single message	Current message	Destination response data
connectorMap	Single connector step	Current step	Inter-step data passing

export function transform(msg: IntuMessage, ctx: IntuContext): IntuMessage {
  let seq = (ctx.channelMap.get("sequence") ?? 0) + 1;
  ctx.channelMap.set("sequence", seq);

  const data = JSON.parse(msg.rawData);
  data.sequenceNumber = seq;

  const facilityName = ctx.globalMap.get("facilityLookup")?.get(data.facilityCode) ?? "Unknown";
  data.facilityName = facilityName;

  msg.content = JSON.stringify(data);
  return msg;
}

Tip

Use globalMap sparingly. For large lookup tables, load them once in a channel’s preprocessor and store a reference in globalMap so every channel can read the same data without repeated I/O.

Destination-Specific Transformers

When a channel has multiple destinations, you can customize the output for each one. Place a transformer file alongside the destination reference in channel.yaml:

destinations:
  - ref: fhir-server
    transformer:
      entrypoint: src/channels/my-channel/fhir-transformer.ts
  - ref: data-warehouse
    transformer:
      entrypoint: src/channels/my-channel/warehouse-transformer.ts

Each destination transformer receives the output of the main channel transformer and can reshape it for that specific target:

// fhir-transformer.ts — wrap in a FHIR Bundle for the FHIR server
export function transform(msg: IntuMessage, ctx: IntuContext): IntuMessage {
  const resource = JSON.parse(msg.content);
  msg.content = JSON.stringify({
    resourceType: "Bundle",
    type: "transaction",
    entry: [{ resource, request: { method: "PUT", url: `${resource.resourceType}/${resource.id}` } }],
  });
  return msg;
}

// warehouse-transformer.ts — flatten to a row for the data warehouse
export function transform(msg: IntuMessage, ctx: IntuContext): IntuMessage {
  const patient = JSON.parse(msg.content);
  msg.content = JSON.stringify({
    mrn: patient.identifier?.[0]?.value,
    last_name: patient.name?.[0]?.family,
    first_name: patient.name?.[0]?.given?.[0],
    dob: patient.birthDate,
    gender: patient.gender,
    loaded_at: new Date().toISOString(),
  });
  return msg;
}

Response Transformers

Response transformers process the reply from a destination after delivery. They are useful for extracting IDs, logging acknowledgements, or feeding data back through responseMap.

destinations:
  - ref: fhir-server
    response_transformer:
      entrypoint: src/channels/my-channel/response-handler.ts

response-handler.ts

export function transform(msg: IntuMessage, ctx: IntuContext): IntuMessage {
  const response = JSON.parse(msg.rawData);

  ctx.responseMap.set("fhirId", response.id);
  ctx.responseMap.set("fhirStatus", response.meta?.versionId);
  ctx.logger.info(`FHIR resource created: ${response.resourceType}/${response.id}`);

  return msg;
}

Preprocessors and Postprocessors

Preprocessors run before the main transformer. Postprocessors run after all destinations have been written. Both use the same function signature.

transformer:
  entrypoint: src/channels/my-channel/transformer.ts
  preprocessor: src/channels/my-channel/preprocessor.ts
  postprocessor: src/channels/my-channel/postprocessor.ts

// preprocessor.ts — strip BOM and normalize line endings before transformation
export function transform(msg: IntuMessage, ctx: IntuContext): IntuMessage {
  let raw = msg.rawData;
  if (raw.charCodeAt(0) === 0xfeff) raw = raw.slice(1);
  raw = raw.replace(/\r\n/g, "\r");
  msg.rawData = raw;
  msg.content = raw;
  return msg;
}

// postprocessor.ts — record metrics after all destinations complete
export function transform(msg: IntuMessage, ctx: IntuContext): IntuMessage {
  const fhirId = ctx.responseMap.get("fhirId");
  ctx.logger.info(`Pipeline complete — FHIR ID: ${fhirId}, message: ${msg.messageId}`);
  ctx.channelMap.set("lastProcessedAt", new Date().toISOString());
  return msg;
}

Error Handling

Throwing an error inside a transformer stops the pipeline for that message. The message is routed to error handling (dead-letter queue, retry, or alert) based on your channel configuration.

export function transform(msg: IntuMessage, ctx: IntuContext): IntuMessage {
  const data = JSON.parse(msg.rawData);

  if (!data.patient_id) {
    throw new Error("Missing required field: patient_id");
  }

  if (!data.mrn?.match(/^MRN-\d{8}$/)) {
    throw new Error(`Invalid MRN format: ${data.mrn}`);
  }

  msg.content = JSON.stringify(data);
  return msg;
}

Caution

Unhandled exceptions crash the pipeline for that specific message only — other messages and channels continue processing normally. Always validate critical fields early and throw descriptive errors so the dead-letter queue contains actionable context.

Code Template Libraries

Shared logic can be extracted into code template libraries and reused across channels. Define templates in a code_templates directory and reference them:

my-project/
├── src/
│   ├── code_templates/
│   │   ├── hl7-helpers.ts
│   │   └── fhir-builders.ts
│   └── channels/
│       ├── channel-a/
│       │   └── transformer.ts   ← imports from code_templates
│       └── channel-b/
│           └── transformer.ts   ← imports from code_templates

src/code_templates/fhir-builders.ts

export function buildPatient(mrn: string, family: string, given: string) {
  return {
    resourceType: "Patient",
    identifier: [{ system: "http://hospital.example/mrn", value: mrn }],
    name: [{ family, given: [given] }],
  };
}

export function buildObservation(
  patientRef: string,
  loincCode: string,
  value: number,
  unit: string
) {
  return {
    resourceType: "Observation",
    status: "final",
    code: { coding: [{ system: "http://loinc.org", code: loincCode }] },
    subject: { reference: patientRef },
    valueQuantity: { value, unit, system: "http://unitsofmeasure.org", code: unit },
  };
}

src/channels/channel-a/transformer.ts

import { buildPatient } from "../../code_templates/fhir-builders";
import { Hl7Message } from "node-hl7-client";

export function transform(msg: IntuMessage, ctx: IntuContext): IntuMessage {
  const hl7 = new Hl7Message(msg.rawData);
  const patient = buildPatient(
    hl7.get("PID.3.1"),
    hl7.get("PID.5.1"),
    hl7.get("PID.5.2")
  );
  msg.content = JSON.stringify(patient);
  return msg;
}

Tip

Code template libraries are a great place to put mapping tables, FHIR resource builders, HL7 segment helpers, and any logic that more than one channel needs. Keep them well-typed so every consumer gets autocomplete.