Industry Solutions
Published 2026-05-14Updated 2026-07-013 min read
Industry-validated PoC patterns and demo assets. While other sections of the Engineering Playbook focus on "how to build it (How)," this section focuses on "what value to show customers and how (What/Why)."
Categories
Retail
Comprehensive e-commerce, large-scale offline retail, household goods, beauty, beverages manufacturing-distribution, and other retail industry domains.
- LG H&H Marketing Innovation PoC — Integrated marketing innovation PoC across 3 BUs (Beauty + HDB + Refreshment). First-party data + 4 external signals × Ontology + Agentic AI 8-scenario demo.
Energy (Planned)
Refining, gas station network memberships, and campaigns.
Financial Services (Planned)
Integrated PoC patterns for banking, cards, and securities data.
Manufacturing (Planned)
Manufacturing OEE, SCM, and customer integration.
Common PoC Patterns
All Industry Solutions leverage the following shared assets:
- Knowledge Graph (Neptune, openCypher) — 25-class domain model
- Hybrid Search (OpenSearch + Cohere embed/rerank + Bedrock) — BM25 + KNN + RRF + Rerank
- Persona Switcher — Same data, recomposed through each department's lens
- Agentic AI (Bedrock Sonnet 4.6 + AgentCore) — Autonomous tool invocation + Memory + Code Interpreter
- Cohort Separation — Real data + synthetic data + external signals separated via
cohort_tag - Bedrock Guardrails — 4 topics + marketing consent + PII + industry-specific guards
Use Cases
- Demo assets for customer meetings (30-minute demo path)
- Industry-specific references for PoC proposals
- Solution architecture (SA) sales enablement material
- Pattern reuse when entering new industries
Design Principles
| Principle | Meaning |
|---|---|
| Public-information based | No dependency on customer-confidential materials — anyone can verify |
| Working demos, not concepts | Real data + synthetic + external live signals combined |
| 7–10 scenarios | Not too many; each scenario specifies its data mix |
| 5 department personas | Recomposing the same data through different lenses — the core differentiator |
| Single architecture suitable for an 8-week PoC | Decision trade-offs are separate; the PoC stays simple |