Skip to content
  • Pricing

AI Capabilities

Ilmiya documents three AI surfaces: ILM (Ilmiya Language Model), the institutional AI suite; Aide, the AI assistant portal for end users; and the Canvas recitation evaluator, which scores Quran recitation in real time. All three are controlled at the Organization level.

What it does

Section titled “What it does”

ILM is a suite of AI-driven models that supports teaching through predictive, context-aware automation; administrators manage it under Administration → Intelligence. The ILM dashboard exposes three modules: Experts (expert-style explanations and academic assistance for students), Aide (conversational assistance for staff and students), and Usage Analytics (requests, token usage, and performance trends over time).

Aide is where users work with AI directly. The portal provides text chat with conversation history saved automatically and context carried across each session; file uploads attached to messages for analysis; a real-time voice assistant (documented as RTVI — Realtime Voice Interaction) that answers in both speech and on-screen text; live voice meetings with barge-in, meaning the user can interrupt the AI mid-response; selectable domain-specific AI Experts whose specialization shapes a session; a document mode that generates content into an editable markdown (MDX) editor and exports .doc files; and side-panel tools — calculator, drawing board, text-to-speech, and voice input — that run beside an active chat.

The Canvas recitation evaluator records a learner reciting a verse, transcribes the audio with automatic speech recognition, aligns the transcript word-by-word against the correct text, and identifies exactly which words contained errors and which of four error types applies: Tajweed, Verbal, Waqf, or Memorization. Results render as per-word colour codes (green correct, red error, orange/yellow partial) with a percentage score, and every attempt is saved with score, error count, date and time, and audio playback. Judge mode overlays saved error annotations directly on the Quranic text. Canvas separately includes a voice AI assistant for questions and explanations.

Key objects

Section titled “Key objects”
ObjectDefinition
ILMInstitutional AI suite; deployed self-hosted or third-party hosted
ILM moduleA unit of the ILM dashboard: Experts, Aide, or Usage Analytics
ExpertPre-configured AI assistant tuned to one domain; published per organization
ConversationSaved Aide chat session; title auto-generated from the first message
Live meetingExtended bidirectional audio session with barge-in support
Recitation attemptOne recorded verse with score, error breakdown, timestamp, and audio playback

Configuration

Section titled “Configuration”

Deployment posture is the primary ILM decision: self-hosted models for full institutional data control and privacy, or third-party hosted models for flexibility and scalability. Administrators manage ILM and review its Usage Analytics from Administration → Intelligence.

End-user availability is feature-gated per Organization:

  • Aide as a portal must be enabled by the organization’s administrator.
  • Aide_Live gates live voice meetings. Two meeting types are documented — Gemini Live (Google Gemini) and Azure Live (Microsoft Azure Cognitive Services) — and an organization may have one or both, depending on configuration.
  • Aide_Chat_Speaker gates the Read Aloud (text-to-speech) control on chat responses.
  • Canvas_Recite gates recitation evaluation in Canvas.
  • The Experts catalog users can browse is whatever the administrator has published; requests for additional experts go to the administrator.

Constraints and edge cases

Section titled “Constraints and edge cases”
  • File uploads are capped at 25 MB per file. Supported types: PDF, plain text, PNG, JPG/JPEG, GIF, WebP, and CSV. Files are not stored permanently — to reference the same file in a new conversation, upload it again.
  • Context from voice sessions is not carried into text chat. If a voice connection drops mid-session, disconnect and reconnect to start a fresh session.
  • Live meetings and recitation evaluation stream audio in real time; both degrade on slow or unstable connections, and a recitation evaluation may not complete if the connection drops mid-recording.
  • Recitation accuracy depends on a working microphone and a quiet environment.
  • Deleting a single conversation is permanent; clearing all history deletes every conversation and cannot be undone.
  • Document sessions are stored separately from regular chats, and the editor exports .doc only.
  • Voice input (dictation) transcribes a single message into the input field; it is distinct from the voice assistant’s full spoken conversation.

Works with

Section titled “Works with”