Plan permission-aware AI integration for a retail POS SaaS
Project brief
Commission an architecture assessment for a cloud retail/POS product spanning sales, inventory, customers, online orders and reporting. This engagement produces a plan, not an implementation.
Cover store-sales questions, kitchen-printer support and reorder guidance. Specify documentation/customer-knowledge RAG, ingestion, grounding, MCP tools, existing APIs and CRM analytics. Map support and customer-facing agent workflows with controlled actions and tenant/role boundaries.
Compare self-hosted Qwen with hosted OpenAI/Claude, considering quality, latency, cost and maintenance. Address Python/FastAPI, LangChain or alternatives, SQL and planned ClickHouse analytics; explain whether reorder guidance needs ML/scikit-learn or existing rules.
Authorized architecture/API documentation and sanitized examples must be agreed before work; no access is implied. Please describe relevant architecture experience.
Deliverables & acceptance
What you'll deliver
- Architecture diagram and technology trade-off report.
- Prioritized MVP roadmap with dependencies, risks and effort estimates; evaluation plan.
What the result must meet
- Trace each use case through data, models, tools and authorization boundaries.
- Specify test cases for grounding, forbidden cross-tenant access, latency and cost; label assumptions and unresolved inputs.