<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom">
<channel><title>AI-Enabled RMA — Trust is earned, not given</title><link>https://www.bobhuang.com/series/ai-enabled-rma/</link><description>An AI-assisted returns-and-repairs system where the AI advises and a deterministic policy decides: the rule evaluator, retrieval and scope gating, the MCP tool server, and the wizard end to end.</description><language>en</language><lastBuildDate>29 Sep 2026 12:00:00 GMT</lastBuildDate><atom:link href="https://www.bobhuang.com/series/ai-enabled-rma/feed.xml" rel="self" type="application/rss+xml"/><item><title>AI-Enabled RMA, part 6: The wizard end to end — state, paid repairs, and the tests</title><link>https://www.bobhuang.com/blog/985/ai-enabled-rma-part-6-wizard-end-to-end-tests/</link><guid>https://www.bobhuang.com/blog/985/ai-enabled-rma-part-6-wizard-end-to-end-tests/</guid><pubDate>29 Sep 2026 12:00:00 GMT</pubDate><description>The customer-facing return wizard: six steps addressed by an unguessable id, a state machine with a partial-success outcome, persisted exclusions, a global PostgreSQL sequence for RMA numbers, and the test suite that proves the steps are actually wired together.</description></item><item><title>AI-Enabled RMA, part 5: One policy, three hosts — the MCP tool server</title><link>https://www.bobhuang.com/blog/984/ai-enabled-rma-part-5-one-policy-three-hosts-mcp/</link><guid>https://www.bobhuang.com/blog/984/ai-enabled-rma-part-5-one-policy-three-hosts-mcp/</guid><pubDate>10 Sep 2026 12:00:00 GMT</pubDate><description>AIEnabledRMA exposes the same deterministic policy to an agent over MCP. Read-only tools, narrow projections, a required date so answers are reproducible, and the parameter that is deliberately missing.</description></item><item><title>AI-Enabled RMA, part 4: Retrieval and the lockdown — BM25, scope gating, and untrusted output</title><link>https://www.bobhuang.com/blog/983/ai-enabled-rma-part-4-retrieval-scope-lockdown/</link><guid>https://www.bobhuang.com/blog/983/ai-enabled-rma-part-4-retrieval-scope-lockdown/</guid><pubDate>20 Aug 2026 12:00:00 GMT</pubDate><description>An in-process knowledge index with no embeddings, a scope gate that runs before the model is ever called, a closed JSON contract, citation verification, and a fail-closed fallback. The machinery that makes a language model safe to put in front of a customer.</description></item><item><title>AI-Enabled RMA, part 3: The policy evaluator — nine rules, a rule trace, and the money</title><link>https://www.bobhuang.com/blog/982/ai-enabled-rma-part-3-policy-evaluator-nine-rules/</link><guid>https://www.bobhuang.com/blog/982/ai-enabled-rma-part-3-policy-evaluator-nine-rules/</guid><pubDate>06 Aug 2026 12:00:00 GMT</pubDate><description>Inside RmaPolicyEvaluator: the closed request record, the ordered rules and why the order is the design, how warranty and grace periods are resolved, where repair prices live, and why pricing is per product rather than per warranty tier.</description></item><item><title>AI-Enabled RMA, part 2: Eight projects, one dependency rule</title><link>https://www.bobhuang.com/blog/981/ai-enabled-rma-part-2-solution-layout-eight-projects/</link><guid>https://www.bobhuang.com/blog/981/ai-enabled-rma-part-2-solution-layout-eight-projects/</guid><pubDate>23 Jul 2026 12:00:00 GMT</pubDate><description>The solution layout behind AIEnabledRMA: Domain with no I/O, PostgreSQL data access via EF Core 10, a swappable AI provider, an in-process retrieval index, an MCP server, an MVC wizard, a database admin tool, and the single rule that keeps them from drifting apart.</description></item><item><title>AI-Enabled RMA, part 1: The AI advises, the policy decides</title><link>https://www.bobhuang.com/blog/980/ai-enabled-rma-part-1-ai-advises-policy-decides/</link><guid>https://www.bobhuang.com/blog/980/ai-enabled-rma-part-1-ai-advises-policy-decides/</guid><pubDate>09 Jul 2026 12:00:00 GMT</pubDate><description>An end-to-end returns system where a language model helps the customer describe and troubleshoot a fault, and a deterministic, auditable policy decides eligibility, coverage and money. This is the design philosophy behind AIEnabledRMA.</description></item></channel></rss>