AI Interface Builder (AskISC): Generate full Productions from specs or Excel

Problem

• Implementing new interfaces still takes many hours: picking the right adapters, scaffolding Business Service/Process/Operation classes, authoring Rule Sets, and writing DTLs. Specs often live in Excel, emails, or PDFs and must be translated by hand.

Proposal

• Add an AI‑assisted “Create Interface” workflow in the Management Portal. The chatbot ingests an interface specification (chat text, sample messages, or an Excel template) and generates a ready‑to‑run Production interface: Business Service, Business Process (BPL or code), Business Operation, Rule Set, and DTL(s), plus unit tests and documentation.

What the assistant can create

• Production scaffold (new or added to an existing Production).

• Business Service with the correct adapter and configuration (e.g., HL7 TCP, HTTP, File, SFTP, JDBC, FHIR).

• Business Process (BPL or code) implementing orchestration, branching, retries, and error handling.

• Business Operation with the appropriate outbound adapter and settings.

• Rule Set (Business Rules) for routing and conditional logic.

• DTL transformations between source and target structures (HL7, FHIR, JSON, XML, custom classes).

• Test harness and unit tests (message samples, synthetic data generators).

• Inline documentation and a README summarizing endpoints, mappings, rules, retries, and security.

Inputs supported

• Guided chat: describe the source/target systems, protocols, message types, routing conditions, SLAs.

• Sample messages/files: HL7, FHIR JSON, XML, CSV, flat files.

• Excel interface specification (XLSX/CSV) using a standard template with these sheets/columns:

  • Endpoints: Name, Direction (In/Out), Protocol (HL7 TCP/HTTP/File/DB/etc), Host/Port/Path, Format, Auth Method, TLS, Retry Policy.

  • MessageTypes: Source Standard (HL7 2.x, FHIR, JSON), Event/Resource, Version, Expected Volume, Max Size.

  • Mapping: SourcePath (HL7 PID-5.1, JSONPath, XPath, or property path), TargetClass/TargetPath, DataType, Transform (function/lookup), Required, Default, Conditional (boolean expression), Notes, Example.

  • RoutingRules: Condition, Destination (Operation/Process), Priority, OnError Action.

  • NonFunctional: SLA, Idempotency, Max Retries/Backoff, Audit requirements, PII handling.

• Optional: a zip of lookup tables or code lists referenced in the Transform column.

How it works (flow)

1. User opens Interoperability > Productions and clicks “AI Interface Builder.”

2. Choose a namespace and import an Excel spec or start with chat.

3. The assistant parses the spec, asks clarifying questions, and proposes:

   • Adapters for Service/Operation

   • Namespaces, package/class naming

   • Rule logic

   • DTL mappings inferred from schemas and examples

4. It generates a draft Production and classes in a sandbox (dev) namespace:

   • Package.InterfaceX.Service/Process/Operation

   • Package.InterfaceX.DTL.SourceToTarget

   • Package.InterfaceX.RuleSet

   • Tests and sample messages

5. Preview diff and diagrams (flow, mapping). The user approves, edits, or asks for changes in chat.

6. On approve, it compiles, deploys the Production items (disabled by default), and offers a guided test run with synthetic or sample data.

7. Export as an installable module or deploy via CI/CD.

Capabilities in detail

• DTL authoring: suggests direct mappings, lookups, datatype coercions, code tables, and custom transform functions; validates against schemas.

• Rule authoring: creates conditions from the Excel “RoutingRules” sheet and from natural language prompts (e.g., “If PID-8 = F route to OB_FEMALE; else OB_DEFAULT”).

• Orchestration: generates BPL steps for enrichment, synchronous calls, retry/backoff, compensations, and exception paths.

• Protocol adapters: picks and configures EnsLib.* adapters (HL7 TCP/File/HTTP/DB/FHIR), leaving credentials as secrets/placeholders.

• SQL and code generation: creates helper classes, persistent objects, or operations that call external REST/SOAP/DB endpoints.

• Test automation: creates unit tests that feed example messages and assert routing, field transforms, and operation calls.

Safety, governance, and privacy

• Works only in selected namespaces; default to a sandbox and disabled items until approved.

• No secrets sent to the model; credentials remain placeholders or referenced via a secret store.

• Role‑aware: only users with appropriate privileges can generate or deploy.

• Full audit trail of prompts, generated assets, approvals, and deployments.

• Redaction and data minimization options; on‑prem or customer‑hosted LLM endpoints supported.

MVP acceptance criteria

• Import Excel spec, parse Mapping/Rules/Endpoints sheets, and generate:

  • One Service, one Process (BPL), one Operation, one Rule Set, and at least one DTL.

• Compile and register items into a new or existing Production, disabled by default.

• Provide a preview with editable artifacts and a “Regenerate” option in chat.

• Offer a test run using 1–3 sample messages; show trace with success/failure.

• Export all generated assets (classes, rule sets, DTLs, tests, README) as a package.

Example prompts users could run

• “Create an inbound HL7 ADT^A01 over TCP and map to a FHIR Patient; route pediatric patients to OP_PEDS.”

• “Import this Excel and build the interface; retries 3x with exponential backoff; send errors to EV_Alert.”

• “Generate a DTL for PID to Patient.name and identifiers; add a lookup for MRN system.”

Success metrics

• 60–80% reduction in time to first working interface.

• Fewer mapping/route defects discovered after go‑live.

• Higher new‑developer productivity and faster onboarding.