KAIMo Integration in LLARS - Effort Analysis¶

Initial Situation¶

KAIMo (AI assistance in child protection) is a demonstrator/training tool for professionals dealing with child welfare risk. The current prototype is a static Next.js app with a hard-coded case vignette ("Malaika").

Goal: KAIMo should evolve into a training tool for education and professional development, with the ability to create additional case vignettes.

Integration: KAIMo is integrated as a tile in LLARS - similar to existing features like mail rating, LLM-as-Judge, RAG pipeline, etc.

Status (2026-02)

Variant B (LLARS integration) has been implemented. Admin and user panels are production-ready. The following variant sections remain as a historical effort assessment.

KAIMo Scope (prototype)¶

The demonstrator consists of three core functions:

Case file/documents - display of notes and documents
Hint assignment - assign hints to categories:
Basic care of the child
Developmental situation
Family situation
Parents/guardians
Case assessment - risk/resource/unclear + final verdict

AI functions (planned in prototype, but static): - Hint summary - Consequence assessment - Plausibility check

Case 1: Manual HTML workflow¶

Description¶

Instructors create vignettes and texts (summary, consequences, plausibility) in a Word template. A technical staff member inserts them into the demonstrator HTML.

Tasks¶

Task	Description	Effort
Start page extension	New Vue component with case selection tiles	4-8h
LLARS integration	New route `/kaimo`, navigation, permission `feature:kaimo:view`	2-4h
Document structure	Define JSON schema for vignettes	2-4h
Static hosting	Store JSON files in frontend	1-2h
Case loader	Component to load and render JSON	4-8h
Styling	Integrate KAIMo design into LLARS theme	4-8h
Documentation	Word template for instructors	2-4h

Total effort (Case 1)¶

19-38 hours (approx. 2.5-5 days)

Pros¶

Fastest implementation
No backend changes
Works immediately

Cons¶

Technical know-how needed for HTML/JSON
Error-prone manual maintenance
No central management
No versioning of cases

Case 2: LLARS-based case management¶

Description¶

Instructors manage cases and all texts via a LLARS admin UI. Data is stored in MariaDB and shown automatically in KAIMo.

Tasks¶

Task	Description	Effort
Backend
DB schema	Tables: `kaimo_cases`, `kaimo_documents`, `kaimo_hints`, `kaimo_categories`	4-8h
API routes	CRUD for cases, documents, hints	8-16h
Permission integration	`feature:kaimo:view`, `feature:kaimo:edit`, `admin:kaimo:manage`	2-4h

Frontend - Admin
Case editor	Create/edit cases	8-16h
Document editor	Rich-text editor for notes	8-12h
Hint editor	Categorize and assign hints	6-10h
AI text editor	Fields for summary/consequences/plausibility	4-8h

Frontend - Application
KAIMo start page	Case selection with preview	4-8h
Case view	Dynamic rendering from DB data	8-12h
Hint interaction	Drag&drop, categorization, rating	8-16h
Result storage	Store user assessments in DB	4-8h

Integration
LLARS navigation	Tile on home dashboard	2-4h
Styling/Theming	Dark/light mode compatibility	4-8h
Testing	Unit + integration tests	8-12h

Total effort (Case 2)¶

78-142 hours (approx. 10-18 days)

Pros¶

Central case management
No technical knowledge needed for instructors
Versioning possible
User tracking (who assessed which case)
Consistent UX with LLARS

Cons¶

Higher implementation effort
AI texts still manual

Case 3: LLARS + AI-generated content¶

Description¶

Like Case 2, but AI generates the texts (summary, consequences, plausibility) and experts can edit them.

Additional tasks (on top of Case 2)¶

Task	Description	Effort
AI integration
Prompt engineering	Prompts for summary/consequences/plausibility	8-16h
LLM service	Integration via LiteLLM/Mistral (already in LLARS)	4-8h
Generation API	Endpoint `/api/kaimo/generate/{type}`	4-8h
Streaming support	Live output via Socket.IO	4-8h

Frontend - AI
Generation UI	"Generate with AI" button + loading	4-6h
Streaming UI	Live text rendering	4-6h
Edit workflow	Review and save generated texts	4-8h
Regenerate	Re-generate text	2-4h

Quality
Prompt optimization	Iterative improvement	8-16h
Validation	Domain review of AI texts	External
Fallback handling	LLM errors	2-4h

Total effort (Case 3)¶

Case 2 + 44-84 hours = 122-226 hours (approx. 15-28 days)

Pros¶

Maximum automation
Consistent text quality
Fast case creation
Reuse of LLARS LLM infrastructure

Cons¶

Highest implementation effort
AI text quality must be validated
Dependency on LLM availability
Ethical considerations in child protection context

LLARS Architecture for KAIMo (as of 2026-02)¶

Current integration¶

LLARS Home Dashboard
└── KAIMo
    ├── /kaimo            - Hub / entry
    ├── /kaimo/panel      - Case overview
    ├── /kaimo/new        - New case (Admin)
    ├── /kaimo/edit/:id   - Edit case (Admin)
    └── /kaimo/:id        - Case work (Evaluator)

Technical integration (current)¶

Backend (Flask):

app/routes/kaimo/
├── kaimo_admin_routes.py     # Admin API
├── kaimo_user_routes.py      # User API
└── __init__.py

app/services/kaimo/
├── kaimo_case_service.py
├── kaimo_document_service.py
├── kaimo_hint_service.py
├── kaimo_category_service.py
└── kaimo_export_service.py

Frontend (Vue 3):

llars-frontend/src/components/Kaimo/
├── KaimoHub.vue
├── KaimoPanel.vue
├── KaimoNewCase.vue
├── KaimoCaseEditor.vue
├── KaimoCase.vue
├── KaimoAssessmentView.vue
└── KaimoDocumentsView.vue

Database (MariaDB):

kaimo_cases
kaimo_documents
kaimo_hints
kaimo_categories
kaimo_subcategories
kaimo_case_categories
kaimo_user_assessments
kaimo_hint_assignments
kaimo_case_shares
kaimo_ai_content

Permissions:

feature:kaimo:view       -- View cases
feature:kaimo:edit       -- Submit assessments
admin:kaimo:manage       -- Create/edit cases
admin:kaimo:results      -- View results

Recommendation¶

Scenario	Recommended case	Reason
Quick prototype	Case 1	Minimal effort, fast demo
Production use	Case 2	Good balance of effort and benefit
Long-term vision	Case 3	Maximum automation

Recommended approach: 1. Start with Case 1 for a quick demo 2. Develop Case 2 backend in parallel 3. Add Case 3 later as an extension

Effort Summary¶

Case	Effort (Hours)	Effort (Days)	Complexity
Case 1	19-38h	2.5-5 days	Low
Case 2	78-142h	10-18 days	Medium
Case 3	122-226h	15-28 days	High

Note: Effort estimates assume an experienced developer. Testing, reviews, and deployment time are partially included.

Created: November 25, 2025 Author: Claude Code Project: LLARS v2.2