Open to Senior · Staff · Principal

Dhiraj Kumar
Sharma

Senior Software Engineer
Microsoft AI — Bing Ads, DPaaS

I build systems that coordinate work at scale — fleets of AI agents that decompose, fan out, and verify before they finish.

Get in touch Résumé

GitHub LinkedIn Seattle, WA

Armada

Multi-agent orchestration · Microsoft AI, productized independently

A task decomposes, fans out to agents, and those agents fan out again — nesting as deep as the work needs. Results reconcile dynamically level by level rather than only at the end, one branch owning any shared resource, until a single result passes the verification gate. Shipped on GitHub Copilot, Copilot CLI, and VS Code from one canonical core.

AI systems I've shipped

Armada Multi-agent orchestration — nested fan-out where agents recursively decompose further, dynamic reconciliation that folds results back level by level rather than only at the end, single-owner resource scoping, and verification gates before completion.
VS Code extension @armada chat participant, fleet tree, agent/task metadata, language-model tooling, Fleet Hub dashboard — TypeScript.
Copilot CLI plugins /armada fan-out skill with MCP-based fleet observability. /ado-pr for governed Azure DevOps PR approval — two-phase approval, short-lived action tokens, guarded bypass. /ground-rule for persistent AI operating standards.
Agent economics Resource-aware concurrency, lifecycle policies, model right-sizing; local-first fleet state with sanitized summaries and path redaction.
Platform toolkit 29 Terraform/CDKTF skills — security baselines, provider governance, drift resolution, cost estimation, PR quality gates.

3Copilot surfaces
6M/minevents at 99.99%+
$14Minfra removed
10+years shipping

Experience

MicrosoftJun 2026 — now

Senior SWE, Microsoft AI. Armada; leads DPaaS Hadoop/Spark migration to Azure — networking with Private Link, Vault → Key Vault with drift reconciliation, AKS on ephemeral per-job VMs.
Amazon2019 — 2026

Senior SDE. Tier-1 push notifications at ~6M events/min, 99.99%+ availability. Alexa in-app purchase 0 → ~$4M/yr. Hey Disney integration. ~$14M/yr cut from the Rekognition fleet.
Oracle2017 — 2019

Java/Spring Boot microservices; production errors down ~45% via JVM profiling and query tuning. Technical lead for 10+ engineers.
Care.com2016 — 2017

Acquisition funnels and instrumentation — click-through +25%, conversion +21%.
Samsung2016

Artik marketplace features and partner portals — ~27% revenue growth.

Also built

Beampersonal

Reverse-engineered a discontinued Meta Portal TV — no docs, no root, no app store — and shipped a full Android TV launcher solo. 700+ live channels, 11 languages, 25+ sideloaded apps. Daily driver.
Job-search automationpersonal

Built on the open-source career-ops pipeline (santifer/career-ops, MIT): scheduled board scanner, a résumé generator that edits the master .docx in place so styling stays byte-identical, and a submit flow that always stops for human review.
DPaaS → AzureMicrosoft

Multi-region Hadoop/HDFS and Spark platform where all Bing Ads bidding data lands, moving out of the datacenter onto Azure.

Stack

Multi-agent orchestrationMCPCopilot extensibility TypeScriptPythonJavaKotlinReact AzureAKSKey VaultTerraformCDKTF AWSLambdaDynamoDBSQS/SNS KubernetesDockerSparkHadoop

M.S. CS, Texas A&M–Kingsville · Amazon Security Certifier · RHCE · GKE Architecting · 40+ open-source workshops taught across India

Dhiraj KumarSharma

Dhiraj Kumar
Sharma