Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.kaireonai.com/llms.txt

Use this file to discover all available pages before exploring further.

This page captures the forward-looking roadmap for KaireonAI. It is not a delivery contract — dates are deliberately omitted and priorities shift as we learn from pilot deployments. Near-term items are what the team is actively working on; mid-term is queued; parked items are intentionally deferred until we see pull for them. For what’s already shipped, see the changelog.

Near-term (active backlog)

Work either in flight or next up. Everything here is gated by its own success criteria rather than a fixed date.

Fairness hard-gate (block flow activation on disparate impact)

Today the platform reports fairness metrics — /api/v1/fairness/evaluate computes disparate impact, equal opportunity, demographic parity, and counterfactual fairness; /api/v1/fairness/report returns historical trends. Operators see the numbers but nothing prevents a decision flow from going live when those metrics breach a configured threshold. Public-sector, healthcare, and regulated-finance personas (credit, underwriting) cannot deploy without this hard-block. Intended scope:
  • Per-tenant fairness policy (configurable thresholds per metric + per protected-attribute set) stored alongside the existing ranking profile so policy changes ship through the same governance flow.
  • Block-on-publish enforcement in decision-flows/[id]/publish — evaluate the policy against the latest decision-trace sample; if any threshold is breached, return 422 with a structured violation report and refuse to mark the flow published.
  • Continuous re-evaluation via a scheduled fairness-check cron that flips a published flow back to paused if metrics drift past the threshold post-launch (operator opt-in).
  • Override workflow — an explicit four-eyes governance approval (re-using the existing approvals workflow) that lets a documented human override unblock publish or keep-live for cases where the threshold breach is justified.
Surfaced by the 2026-05-05 persona audit; gating prerequisite for public-sector and healthcare deployments.

Dedicated /api/v1/ai/explain HTTP route

The Executive Dashboard’s Explain button on each anomaly row today falls back to a deterministic summary built from the anomaly tuple’s fields. The full LLM-narrated explanation pipeline is already implemented internally but is not yet exposed as a public HTTP endpoint. Wiring up /api/v1/ai/explain unlocks richer, causally-framed explanations on demand for both the dashboard and external integrations.

Pilot guardrails for reports + alerts

A set of safety rails to ship before enabling EventBridge automation by default:
  • Per-tenant report-schedule cap — prevent a single tenant from scheduling 50 daily LLM-narrated reports.
  • Minimum schedule cadence — reject cron expressions that would fire more often than once per hour.
  • Default LLM-narrative opt-out — new schedules default to narrative-disabled; operators explicitly opt in.
  • Per-tenant monthly LLM spend cap — track narrator invocations against a budget; degrade gracefully to narrativeless reports when exceeded.

EventBridge automation wire-up

Once guardrails are in place, move alerts and report schedules from on-demand (Run Now, manual tick) to automated background evaluation via AWS EventBridge → /api/cron/tick. See EventBridge Setup for the mechanics; the wire-up is optional during pilot and will remain optional for self-hosted deployments.

S3 artifact storage for large report runs

Today, ReportRun.artifactPayloads stores artifacts inline as base64-encoded JSON. This is fine for small PDFs and CSVs but becomes a database-bloat risk for large runs. Moving artifacts to an S3 bucket (with a signed-URL download path) keeps the database lean and handles arbitrary payload sizes. Gated on a planned REPORTS_S3_BUCKET env var; inline mode remains the default.

Mid-term

Queued work. Not in progress yet, but sequenced to land after the near-term items.

Slack / Teams interactive components

Approve / dismiss buttons directly in the notification payload so on-call engineers can acknowledge an alert or mark a report as reviewed without leaving the conversation. Today both adapters render read-only messages.

Scheduled screenshot-style dashboard PDFs

Current exports are data-driven — the runner regenerates the report from the underlying data sources at run time. Complement this with layout-capture PDFs that mirror the dashboard’s rendered UI pixel-for-pixel. Useful for executive packs where the rendered visualisation is the deliverable.

Custom dashboard builder

A drag-and-drop dashboard composer where operators pick widgets (KPI card, time-series chart, table, heatmap) from a palette and bind each to a data source. Closes the loop started by the existing dashboard Export and Save as Report affordances — any custom dashboard instantly becomes a scheduled report. Time-limited, rotatable URLs that render a read-only dashboard (or report run) without requiring a login. Useful for ad-hoc external sharing (board decks, investor updates) without provisioning new user accounts.

Terraform / CDK modules for EventBridge and App Runner

Codify the manual AWS Console setup documented in EventBridge Setup as reusable IaC modules. Covers the scheduler rule, IAM role, Parameter Store entry, and App Runner environment-variable wiring so a new pilot can promote to fully automated in one terraform apply.

Parked / future

Intentionally deferred. Listed so there’s no ambiguity about what we’re not working on — if any of these become a hard blocker for a pilot, let us know and we’ll re-prioritise.

Mobile SDKs

Native iOS / Android SDKs for client-side Recommend + Respond. Parked until we see meaningful pull from mobile-first pilots; the REST APIs are fully usable from native HTTP clients today.

Shared read-only demo tenant

A single always-on tenant with pre-loaded decisioning scenarios so prospective users can evaluate the platform without creating an account. Parked until user volume on playground.kaireonai.com makes the read-only sandbox cost-effective to maintain.

Full i18n rollout

The i18n foundation is in place — locale message catalogues plus a hook that resolves translated strings — but the per-page wiring across all 54 UI pages has not been completed. Currently ~5% of pages are wired. Parked until we have an identified non-English pilot; the effort is mechanical but sizeable.

Re-enable CI / CodeQL workflows

CI and CodeQL jobs are currently set to manual-trigger only because the repo was recently open-sourced under a squashed-history commit and we don’t want to flood the Actions tab with backfill runs. Re-enabling auto-triggers is a one-line change once the repo is stably public.