Custom AI Model Architecture & Enterprise Integration

High-level architecture of SmartERP's integrated custom AI models, model serving infrastructure, and hybrid cloud-edge topologies.

1. Overview & Technical Architecture#

SmartERP embeds custom enterprise AI models engineered specifically for African commercial operations, Ethiopian statutory compliance, and Ge'ez-based multilingual language processing.

Unlike off-the-shelf public models that lack domain context for Ethiopian accounting proclamations (e.g., Proclamation 979/2016) or commercial terminology (e.g., Model 19 GRN, Model 20 SIV, CPO bid bonds), SmartERP's integrated models are pre-trained and fine-tuned on curated corporate datasets.

Note

Inference Engine vs. Training Pipeline Architecture: SmartERP's native module orchestrates runtime inference, prompt templating, PII sanitization, context window management, and model endpoint health monitoring (via vLLM, Ollama, or REST gateways). Parameter fine-tuning (LoRA / QLoRA) is executed via external offline data engineering pipelines (e.g., PyTorch / Hugging Face on high-performance GPU clusters) using curated datasets, with trained adapter weights subsequently loaded into the inference server.

flowchart LR
    subgraph Enterprise ["SmartERP Monorepo Application"]
        ORM["Odoo 19 ORM / Models"]
        Chat["Floating AI Composer (.mk_voice)"]
        Cron["Event Bus & Crons"]
    end

    subgraph Middleware ["Ashewa AI Bridge Layer"]
        Proxy["PII Redaction & Security Gateway"]
        Cache["Semantic Response Cache (Redis/PGVector)"]
        Router["Dynamic Task Router"]
    end

    subgraph Inference ["Model Serving Infrastructure"]
        Local["On-Premise Sovereign Model (vLLM / Ollama)"]
        Cloud["Enterprise Gemini / Claude Endpoint"]
    end

    ORM --> Proxy
    Chat --> Proxy
    Cron --> Proxy
    Proxy --> Cache
    Cache --> Router
    Router -->|Confidential / Sovereign| Local
    Router -->|High-Reasoning Analytical| Cloud

2. Prerequisites & Infrastructure Requirements#

Runtime Environment: Python 3.12+, PostgreSQL 16 with pgvector extension.
Security Access Group: Ashewa AI / Administrator (Technical ID: muk_ai.group_ai_manager)
Hardware Profile: Minimum 16 GB System RAM (CPU inference) or NVIDIA GPU with 8+ GB VRAM.

3. Step-by-Step Configuration & Verification#

Step 1: Navigating to Model Management#

Access your SmartERP instance at https://smarterp.et/odoo.
Open Settings → Technical → Ashewa AI → Model Registry.
Select the custom model profile.

{screenshot:10-overview-step1-navigation}

Step 2: Testing Endpoint Health & Response Verification#

Click the Run Diagnostic Health Check button.
Verify token generation speed (tokens/sec) and latency benchmarks.
Review sample response in the diagnostic console.

{screenshot:10-overview-step2-verification}

4. Advanced Configuration Points#

Automated Tasks: muk_ai.ir_cron_model_health_monitor runs every 15 minutes.
Multi-Tenant Boundaries: Custom model fine-tuning adapters are strictly segregated per tenant database.

5. Comprehensive Troubleshooting Matrix#

Symptom / Error	Probable Root Cause	Resolution Action
`Connection refused: Model server unreachable`	Inference daemon stopped or port blocked by firewall.	Restart model service (`systemctl restart vllm`) and verify firewall rules.
`Token limit exceeded on input prompt`	Input context exceeded window size.	Enable context truncation in model profile settings.

መጨረሻ የተሻሻለው: 2026-10-30