Kinetic GainSystem Control Plane

RAG Deflection Bot

Knowledge-grounded support deflection for billing, auth, integrations, and workflow operations. This repo models when retrieval can safely answer, when it should partially guide, and when it must escalate immediately.

System Artifact / Principal Technical Spec

Retrieval-Grounded Support Deflection

The rag-deflection-bot repo models support deflection as a governed operating surface. It makes knowledge freshness, escalation boundaries, and risky reply scenarios explicit instead of assuming every customer question should be answered by automation.

Primary purpose

Reduce L1 support load by routing low-risk requests through grounded knowledge answers while preserving fast escalation for billing, identity, integration, and duplication-risk scenarios.

Application shape mapping

src/app.ts serves the operator shell. src/data/sampleDeflection.ts models knowledge and case posture. src/services/ragDeflectionService.ts exposes API-ready payloads. src/services/render.ts renders the control-plane routes.

Commercial interpretation

For GTM leaders, this shows how self-service support can protect gross margin without silently shifting customer risk into unsupported automation. For technical reviewers, it shows the concrete guardrails needed before a RAG assistant touches real customer workflows.

Spec Classification

Target platform
Node.js Web Runtime (Express / HTML diagnostics)

Architecture role
Director of Web & GTM Systems

Signal metric target
89% Signal Clarity

Active focus
Support containment, knowledge freshness, escalation safety, retrieval-grounded answers