Abhijit
Krishna Menon.

My contribution

Platform setup and engineering of an extensible analytics harness, with custom SQL-based tool onboarding. Contributors supplied domain context and SQL tools.

What it supports

MCP integrations, code interpreters, evaluations, intent and sub-topic routing, retrieval over a knowledge base, and safe, access-controlled answers. The React portal and Python backend support charts, reports, and self-service analytics.

Making it usable

Self-service onboarding of business domains, sources, and knowledge articles, alongside authoring, testing, and CI tooling for skills and topics. Systematic evaluations replaced manual log inspection for quality measurement.

Starting point

The program had no analytics around this workstream. I defined metrics, built Redshift models and pipelines, and created reporting that helped operations and BI partners see where excess dispute volume concentrated and what drove it.

Following through

When partner teams lacked capacity for needed product changes, I worked in the customer-facing portal and shipped UI updates and an automated point-of-entry validation gate.

What changed

The work progressed from analysis and recommendations to self-service visibility and a production control. My ownership crossed the data, backend, and frontend boundaries needed to deliver it.

Explore

Built a Redshift MCP server for schema discovery, query execution, and performance diagnostics. Its secure proxy and identity layer support per-agent identity, audit provenance, and permission guardrails.

Review

Built an agent that executes pull-request SQL against Redshift, parses EXPLAIN plans, and surfaces syntax errors and optimization recommendations before merge.

Investigate

Built an autonomous on-call agent for ticket triage, parallel investigations, root-cause analysis, and automatically posted findings.

Further work

A voice-enabled coaching and learning agent turns performance data into spoken briefings. Reusable permission and ticketing frameworks support multiple applications.

Launched at Amazon’s Ignite Live, DSP Companion is publicly described as an AI assistant that helps Delivery Service Partners run their operations, build stronger teams, and improve delivery experiences. I contributed to the product. See the public launch announcement ↗