Kural AI System
AI Interaction Guidelines
AI தொடர்பு வழிகாட்டுதல்கள்
A shared design direction for how Kural AI should behave, respond, and collaborate responsibly alongside human operators.
குறள் AI எவ்வாறு செயல்பட வேண்டும், பதிலளிக்க வேண்டும் மற்றும் மனிதர்களுடன் இணைந்து பணியாற்ற வேண்டும் என்பதற்கான பகிரப்பட்ட வடிவமைப்பு வழிகாட்டுதல்கள்.
அறிமுகம் (Introduction)
Kural AI (குறள் AI) is the cognitive assistant of the Tamil Design System, engineered to support government administration, civic services, and public software development. These guidelines establish the behavior, feedback mechanics, accessibility standards, and ethical guardrails required when weaving artificial intelligence into user workflows.
1. Make Workflows Proactive
பணிப்பாய்வுகளை செயல்திறன் மிக்கதாக மாற்றவும்
AI experiences must demonstrate a deep understanding of the operator's physical and digital context. Instead of sitting idle waiting for manual triggers, Kural AI should analyze active screens, local network status, and operational histories to make smart assumptions and offer proactive aid.
- •Pre-fill complex form sections using verified database attributes.
- •Identify network speed drops and queue actions before offline failure occurs.
DO:Suggest relevant welfare schemes to an administrative officer based on the beneficiary's household income and district metrics.
DON'T: Send frequent pop-up interruptions requesting user inputs before displaying search results.
2. Keep the Flow State
தொடர்ச்சியான பணி நிலையை பராமரிக்கவும்
Preserve the user's focus by keeping assistant controls integrated, natural, and contextually embedded. Avoid throwing modal dialogues or forcing screen swaps that cause cognitive disruption. When users transition between mobile hardware and desktop portals, maintain input histories seamlessly.
- •Use inline widgets and side drawers rather than full-page redirects.
- •Retain multi-page wizard state if a connection drop forces a browser refresh.
DO: Embed AI-driven auto-suggestions within the active input card so they appear directly below the active cursor focal point.
DON'T: Trigger full-screen popups that steal keyboard focus mid-typing and force physical mouse clicks to dismiss.
3. Content and Form are Dynamic
உள்ளடக்கம் மற்றும் வடிவம் மாறும் தன்மை கொண்டவை
Kural AI must accept inputs in whatever format best suits the operator: text, voice, upload files, or visual forms. Dynamic layout components must stretch and wrap automatically to handle the resulting language translation length expansions.
- •Support audio/voice intake for elderly citizens filing welfare applications.
- •Enforce auto-wrapping container boxes to support dynamic Tamil character expansions.
DO: Automatically transition from a rigid grid table into a list configuration when long localized Tamil translation arrays exceed layout boundaries.
DON'T: Truncate long Tamil strings in buttons with fixed widths, causing them to show unreadable trailing dots.
4. Make it Easy to 'Pop the Hood'
கணினி நிலைகளை வெளிப்படையாகக் காட்டவும்
Build institutional trust by offering operators clear insights into the system status and the logic used to produce AI outcomes. Technical logs, decision records, confidence metrics, and source guidelines must be easily retrievable in a single click.
- •Provide hover annotations on key metrics showing exactly what factors were weighted.
- •Provide clear links to active source codes and public welfare rules.
DO: Provide a clean accordion listing the database references and confidence scores used in the automated application routing.
DON'T: Present automated AI classification badges as absolute truths without explaining the underlying reasoning.
5. Promote Collective Intelligence
கூட்டு அறிவாற்றலை ஊக்குவிக்கவும்
Connect individual task streams to the broader organization context and knowledge. Surfaces relevant expertise, shared resources, and resolved histories so operators can learn from past decisions without violating data confidentiality controls.
- •Surface similar resolved grievances when a welfare officer reviews an active case.
- •Automatically index custom operational resolutions to update shared help guidelines.
DO: Suggest pre-written response templates approved by senior team members during grievance resolution.
DON'T: Force designers or developers to start from scratch when similar structural layouts already exist in the shared library.
6. Expand to Multi-Player
பல்பயனர் மற்றும் பல-அறிவாற்றல் நிலைக்கு விரிவுபடுத்தவும்
Support multiplayer collaborations by letting tasks move fluidly between individual users, team members, and intelligent sub-agents. Maintain shared history trees, version indicators, and context parameters across all active collaborators.
- •Allow manual overrides and handoffs between the AI system and supervisors at any time.
- •Annotate changes made by the AI model vs. manual user edits using distinct indicators.
DO: Maintain a clear audit history showing which attributes were modified by which model version and when the human operator approved them.
DON'T: Lock component editing states, preventing supervisors from modifying AI proposals.
7. Weave TamilDS into the Experience
தமிழ் DS அடையாளத்தை ஒருங்கிணைக்கவும்
AI outputs should naturally echo the visual character, design constraints, and cultural tone of the Tamil Design System. Tone, layout spacings, and typography metrics must adhere to the design system regulations.
- •Use standard TamilDS font variables and HSL semantic colors in generated assets.
- •Adopt a respectful, polite, and culturally clear tone in all localized Tamil text outputs.
DO: Generate user interface layouts utilizing saved design tokens, keeping brand consistency intact.
DON'T: Generate layout styles using hardcoded random colors or tight vertical line-height rules that break readability guidelines.
8. Always Accelerate Responsibly
எப்போதும் பொறுப்புடன் துரிதப்படுத்தவும்
Prioritize quality and accuracy over reckless speed. The system must clearly indicate operational limitations, alert users to potential data anomalies, and demand explicit manual validation checks before modifying production systems.
- •Demand manual verification before executing critical data alterations.
- •Provide clear warning labels if AI calculations fallback to default estimations.
DO: Implement clear confirmation dialogue modals with dynamic warning checklists before bulk-updating citizen registry records.
DON'T: Execute bulk actions or update database rows silently without providing undo operations or manual history logs.