V. VINOD LOGANATHAN RAMESH KUMAR
← Selected work

Intuit · Developer support

AI-powered Slack support

Intent routing, grounded knowledge retrieval, and human handoff for cloud support.

Approximate reported outcome ~40% less support triage time for cloud support requests.

What this result covers

Workflow scope
Cloud support requests handled through the AI-powered Slack support workflow at Intuit. The figure describes triage time; it does not establish a reduction in end-to-end resolution time.
Measurement context
An approximate reduction reported in my professional experience. The public summary does not specify the before-and-after periods, baseline triage duration, request count or mix, or the start and stop events used to time triage.
AI-POWERED SLACK SUPPORT
01 Ask in Slack Question · service · symptoms
→
02 Route & ground Intent · runbooks · context
→
03 Guide resolution Verified response · handoff
Confluence knowledge GCP service context Langfuse evaluation
Conceptual overview of my engineering work

Problem

Support requests arrived in Slack without consistent routing, service context, or an immediate path to the right runbook.

My contribution

Engineered an assistant that identifies intent, retrieves relevant knowledge and context, and returns verified guidance before a human handoff when needed.

Scope and ownership

I built the support path from a Slack request through routing, grounded knowledge retrieval, and escalation. Langfuse observability makes the workflow traceable for evaluation.

Architecture decision

I used intent-based routing before retrieval rather than searching every source for every question. A single broad retrieval path was easier to build, but it returned noisier context and made evaluation less precise. Routing first narrowed the source set, improving relevance and making unsupported requests safer to escalate.

ADR 02 · Route intent before retrieval

Portfolio decision summary

Alternative considered

Searching every source for each question is easier to build, but returns noisier context and makes evaluation less precise.

Trade-offs & consequences

Routing narrows the source set. Unsupported intents need an explicit escalation path, and each route needs evaluation.

Operating constraints

Guidance must be grounded in current knowledge, traceable for evaluation, and safe to escalate when the assistant cannot resolve the request.

This case study summarizes my professional work. The diagram is a conceptual overview, and architecture decisions are portfolio summaries.