South Africa SSEG RAG Example

This example shows how to integrate ODS-E metadata with a municipality-aware RAG assistant for SSEG application support.

Problem Setup

Target scenario:

• applicants and municipal officials ask policy and process questions on a shared platform
• each municipality has distinct requirements and tariff/process rules
• national policy context must still be available when local policy is silent

Minimal Metadata Contract

For each indexed chunk, store:

• municipality_id (for example za.wc.city_of_cape_town)
• municipality_name
• role_scope (applicant, official, both)
• tariff_schedule_id (if relevant)
• seller_party_id / buyer_party_id / network_operator_id (for settlement and wheeling context)
• distribution_zone, feeder_id, voltage_level (when topology-specific guidance applies)
• source, section, effective_date

Retrieval Flow

1. Accept request context:
   • user question
   • selected municipality
   • user role
2. Build retrieval filter:
   • strict municipality match first
   • role-compatible chunks
   • tariff/topology constraints if present
3. Backfill with national chunks only when municipality-specific content is insufficient.
4. Generate response with explicit citations and action-oriented next steps.

API Shape (Example)

{
  "message": "Can this customer export from a 150 kW system?",
  "municipality_id": "za.wc.city_of_cape_town",
  "role": "applicant",
  "tariff_schedule_id": "za-city-capetown:cpt:LT-MD-2026:v1",
  "party_context": {
    "seller_party_id": "za-nersa:trader:ETANA-001",
    "network_operator_id": "za-eskom:network_operator:WC-01"
  }
}

Answer Policy

Enforce this response structure:

1. Direct answer for the selected municipality.
2. Conditions/limits (capacity, inverter, process step, tariff implications).
3. Exact source citations (document + section).
4. Concrete next step in the application process.
5. If municipality rule is missing, clearly label fallback as national guidance.

Why ODS-E Improves LLM Reliability

• Canonical IDs remove ambiguity between similarly named entities.
• Tariff and settlement keys constrain generation to the right commercial context.
• Municipality and topology keys prevent cross-municipality leakage.
• Structured metadata supports deterministic filtering before the model generates text.

Related Docs

• LLM Integration Overview
• Energy Timeseries Schema
• Asset Metadata Schema