Context / Problem

Clinicians were spending significant time manually transcribing standardised assessment forms from paper and PDF into digital records. The process was error-prone, created workflow bottlenecks, and made consistent scoring difficult. Assessment instruments vary in structure but share common patterns (sections, question types, scoring rubrics) that lend themselves to a structured, schema-driven approach.

Constraints

• •Clinical environment. Accuracy of scoring and data capture is non-negotiable.
• •Assessment instruments are published as static PDFs with no machine-readable structure.
• •Clinician users include neurodivergent practitioners. UX must minimise cognitive load, reduce visual noise, and support focused task completion.
• •Data governance requirements: clear boundaries on what is stored, where, and for how long.

Solution Approach

Built a three-stage pipeline that decouples ingestion from rendering:

1. 1.PDF analysis. Source PDF instruments are analysed and structured data is extracted into a canonical JSON schema. Each section, question, and scoring rubric is mapped to a normalised data model.
2. 2.Canonical schema. The schema serves as the single source of truth, decoupling the source format from the rendering engine. New instruments can be onboarded by producing a schema. No UI code changes required.
3. 3.Interactive form rendering. Schemas are rendered as interactive digital forms (SurveyJS-based) with real-time scoring, validation, and section-level progress tracking.

Architecture

• •Ingestion layer. PDF analysis produces canonical JSON schemas conforming to a defined instrument specification.
• •Schema registry. Canonical schemas stored with versioning and deterministic identity (content-derived slugs), enabling reproducible builds and eliminating manual ID management.
• •Rendering engine. SurveyJS-based form renderer consumes schemas and produces interactive assessments with real-time scoring.
• •Scoring engine. Automated scoring logic embedded in the schema definition. Subscale calculations, reverse scoring, and threshold flagging are computed in real time as the clinician completes the form.

Automation & Integration

• •Schema generation. AI-assisted extraction of structure from PDF instruments, with manual validation and correction at every stage.
• •Deterministic identity. Each instrument, section, and question receives a content-derived slug. This eliminates brittle numeric IDs and reduces friction when instruments are updated or versioned.
• •Real-time scoring. Scoring rules are defined in the schema and executed client-side as forms are completed. No round-trip to a server required for score calculation.
• •Instrument onboarding. Adding a new assessment instrument requires producing a conforming schema. The rendering engine handles the rest.

Governance & Controls

• •Data boundaries. Assessment responses are processed client-side. The platform defines clear boundaries on what is persisted, what is ephemeral, and what is exportable.
• •Schema validation. Instruments must conform to the canonical specification. Malformed schemas are rejected at onboarding.
• •Auditability. Schema versions, scoring logic, and instrument definitions are versioned and traceable.
• •Human-in-loop. AI-assisted schema extraction is always reviewed and validated by a human before an instrument enters production use. AI accelerates the analysis; clinicians and architects validate the output.

Security

Assessment data is handled with appropriate sensitivity controls. The platform does not store identified patient data. Schema definitions and scoring logic are versioned and integrity-checked. Access controls and data handling follow the principle of least privilege.

Delivery Notes

• •Designed and operated as a productised workflow automation capability, applying the same governance and architecture patterns used in enterprise engagements.
• •Production deployment supporting multiple assessment instrument types.
• •Clinician-facing interface with neurodivergent-first UX: reduced visual clutter, consistent navigation, clear progress indicators, and minimal context-switching.

Outcomes

• •Production platform supporting multiple standardised assessment instruments with automated real-time scoring.
• •Canonical schema approach enables rapid onboarding of new instruments without UI code changes.
• •Neurodivergent-friendly interface validated through clinician feedback during development.
• •Deterministic identity model (content-derived slugs) eliminates a class of data management errors common in form-based systems.

Extensibility / Next Steps

Schema-driven architecture supports additional instrument types, export formats, and integration with clinical record systems. The canonical model is designed to accommodate future requirements (e.g. longitudinal tracking, multi-rater instruments) without structural changes to the rendering or scoring engines.