How it works, end to end
Ingest program materials, scope retrieval to the activity, answer with citations, align each message to learning objectives, and analyze outcomes across programs, sites, and conferences.
The high-level flow
Ingest content → Scope and index the activity → Answer with citations inside the activity → Align messages to objectives with confidence scoring → Analyze outcomes and themes → Export CSV for deeper work.
What sets this apart
- •Activity-scoped retrieval that keeps answers inside accredited materials
- •Citations that open the exact slide or timestamp
- •Message-level alignment to learning objectives
- •Analytics that roll up across programs, sites, and conferences
- •Independence and privacy preserved from design to export
Components
Ingestion and enrichment
Services for slides, videos, transcripts, and PDFs
Retrieval and generation
With activity scoping and citation rules
Objective alignment engine
With confidence scoring and editor review
Exports and SSO
For data access and secure embedding
Ingest and normalize
Upload slides, videos, transcripts, and PDFs. The system parses structure and prepares content for retrieval.
- •Formats: PPT, PDF, videos, transcripts, long-form PDFs and articles
- •Parsing: slide headers, sections, and speaker notes when available
- •Transcript alignment: link citations to slide numbers or timestamps
- •Metadata: program title, faculty, therapeutic area, release and expiry, internal IDs
- •Languages: 50+ supported for ingestion and for answer output
- •Versioning: reprocess updates and keep a diffed history
Isolation and scope
Each tenant, program, and session is scoped by keys. Retrieval cannot cross program boundaries.
- •Tenant isolation across organizations and applications
- •Program-scoped retrieval that restricts search to one accredited activity
- •Allowlists and disclaimers to control domain and compliance language
- •Rate limits and session controls to protect performance
Retrieval and answer generation
Queries stay inside the activity. Answers are grounded in sources and always cite them.
- •Retrieval: activity filter first, then semantic search and re-ranking
- •Guardrails: on-label, guideline-aligned language, and scope enforcement
- •Citations: slide numbers or timestamps are included for factual statements
- •Decline when uncertain: the assistant defers if evidence is insufficient
- •Style controls: tone, depth, bullets or narrative, citation density
Objective alignment for every message
Configure in Admin Console, view analytics in Analytics
Each learner message is classified to one or more learning objectives with a confidence score. Editors can review and override.
- •Objectives catalog: define objectives per activity and map them to content
- •Alignment labels: aligned, unaligned, or ambiguous
- •Confidence scoring: numeric score to focus reviewer effort
- •Review tools: accept, relabel, or mark ambiguous with audit logging
Why it matters
Objective alignment shows how questions cluster around what faculty intended to teach and where gaps may remain.
Analytics and themes
Signals from chat and citations become outcomes analytics for providers.
- •Participation and engagement: unique learners, sessions, questions per session, median dwell time
- •Evidence use: citation opens, reference interactions, slide and section heatmaps
- •Theme mining: clustered question topics mapped to learning objectives
- •Objective coverage: alignment rate and messages per objective with trends
- •Portfolio rollups: compare programs, sites, and conferences over time
Exports and data access
Export message-level data for audits, research, and custom reporting.
Programmatic access is available for customers that prefer pipelines.
Embedding, SSO, and performance
Add the assistant to your site with a secure snippet. Run under SSO when needed.
- •Secure embedding: HMAC-signed initialization with a time-bound signature
- •Domain allowlist and hostname validation
- •SSO support: JWT claims for
sub,iss,exp,iat, andaud - •Performance: lazy loading, resource hints, and a two-phase load
- •Kiosk mode: conference-ready with session persistence and a PIN to exit
Observability and reliability
Operational controls keep ingestion and chat healthy at scale.
- •Health checks and monitors for throughput and latency
- •Centralized logging for ingestion and chat events
- •Alerting on error rates and unusual event patterns
- •Backups and recovery with regular verification
- •Horizontal scaling and index tuning for consistent response times
Privacy, security, and independence
Independence and privacy are enforced at each step.
Independence
Providers control content and configuration; supporters see aggregated, de-identified insights only
Privacy
No learner-level reporting to supporters; pseudonymized user hash for analytics
Security
Encryption in transit and at rest; least privilege access; audit logs for changes
Compliance posture
HIPAA safeguards under BAA when PHI is in scope; controls aligned to SOC 2
Optional module: measured learning outcomes
Enable when you need to claim learning outcomes directly.
- •Pre and post knowledge checks tied to learning objectives
- •Competence scale to track confidence on key tasks
- •Commitment to change at post with a 30 to 60 day follow-up
- •Methods transparency with sample sizes and effect sizes
This module complements objective alignment and analytics. Talk to us about enablement and rollout options.