DIGITAL CANDIDATE / ADMIN
Platform intelligence
Develop your AI. Improve every candidate experience.
RESTRICTED WORKSPACE
One place to tune the platform
Inspect evidence, compare responses, build vector indexes and review usage. Access requires a verified platform super-admin account.
Scope & audience
CONTROL ROOM
A measured path to better answers
YOUR WORKFLOW
From evidence to a reviewed release
- Inspect the sourceCheck titles, sections, versions and publication scope.
- Compare retrievalRun the active baseline and your draft against the same question.
- Review quality and costLabel relevance, inspect citations and record human judgments.
- Promote a profileApply settings to one candidate or the platform; retain prior versions.
CONNECTIONS & RATES
Azure configuration
Credentials and deployment prices are managed in Azure application settings. Unknown rates remain unknown.
Open Azure Cost Management ↗RETRIEVAL LAB
Compare the answer and its evidence
Retrieval
Semantic ranking receives at least 50 vector candidates. Scores indicate rank, not factual confidence.
Evidence & generation
Each action runs once. Generation can make one citation-recovery attempt. Saved run summaries omit question text and evidence; exports include them.
Record a human review
Mark relevant results above. Supply additional known relevant IDs to include missed records in recall.
INDEXING
Shape the evidence before retrieval
Application documentation index
Application documentation is indexed through the existing controlled documentation publisher. Candidate chunk settings do not change that corpus. Publisher instructions ↗Chunk design
Chunks stay within their source record. Overlap must be less than half the chunk size. Changes require a new build. Embeddings include each chunk’s title and content.
Builds for this candidate and audience
Index up to 16 chunks per click. Provider operations must be enabled for indexing or Search cleanup. Activate only after reviewing a completed build; stale builds cannot serve changed evidence.
Delete excludes an inactive build from AI immediately. Continue cleanup until its Search documents are removed. Current source knowledge remains available to AI; manage its eligibility in the candidate workspace.
CONFIGURATION HISTORY
Review, promote, roll back
Save the current draft
The retrieval, chunking and generation controls form an immutable profile. Saving a profile does not activate it.
Saved lab runs
Runs and reviews are test history, not AI evidence. Delete a run to clear its saved reviews too. Spending history is retained. Refresh overview to load recent runs; cleanup is manual.
Recent evaluation records
OBSERVABILITY
Know what each improvement costs
Usage covers recorded application and lab operations since enablement. These are configurable estimates; use Azure Cost Management for your complete invoice. Missing usage or prices never count as a zero-cost operation.