Software Engineer · Distributed Systems

Rakhi
Agrawal.

Software engineer building large-scale distributed systems and data infrastructure. Nearly four years at LinkedIn, building systems that process hundreds of millions of metadata events every day.

Signal · packet

Behind every request, there's a system.

01 · About

Portrait of Rakhi Agrawal

I build the systems that other systems depend on.

I spent nearly four years working on LinkedIn's data infrastructure — the systems that keep track of data, ownership, access, and lifecycle across the company. When it's fast and correct, nobody notices. That's the point.

Off the clock, I build things I want to use. I built a finance app to track expenses, set savings goals, and manage money across currencies — without paying another subscription. Then I built a workout tracker for the same reason: I wanted something that worked the way I wanted it to. The pattern is simple: if I want something, I'll probably try to build it.

02 · Experience

Where I've built.

  1. Software Engineer · LinkedIn

    Mountain View, CA → Toronto, ON

    Jan 2023 — Jun 2026

    Built backend and platform infrastructure for LinkedIn's metadata ecosystem, powering data discovery, governance, and compliance across the company's data lake.

    • Data retention platform

      Designed a centralized retention-policy and enforcement layer for HDFS and Iceberg datasets as part of LinkedIn's storage-reduction initiative. Built lineage-aware policy resolution, retention synchronization, and validation across 30K+ datasets.

      ~$50M projected annual savings
      30K+ datasets
      • HDFS
      • Iceberg
      • Java
      • Airflow

    • Multi-primary MySQL architecture

      Migrated a compliance-critical metadata store from single-primary to multi-primary MySQL under live production traffic. Built dual-read/write, backfill, validation, and cutover mechanisms to complete the migration with zero downtime and no data loss.

      15K read QPS · 10K write QPS
      Zero-downtime cutover
      No data loss
      • MySQL
      • Java
      • gRPC

    • Metadata ingestion reliability

      Rebuilt failure handling for LinkedIn's metadata ingestion pipeline with automated recovery, DLQ replay tooling, health metrics, and alerting — improving ingestion reliability while reducing operational overhead.

      170M+ daily events
      99.99% ingestion success
      • Flink
      • Kafka
      • OpenTelemetry
      • Java

    • Samza → Flink migration

      Led migration of five critical metadata ingestion pipelines from Samza to Flink and drove production rollout across large-scale streaming infrastructure.

      250M+ records/day
      500+ Kafka topics
      5 pipelines
      • Flink
      • Kafka
      • Samza

    • Query & schema performance

      Cut high-volume query latency from 4–5s to 0.05s through indexing and application-level tuning — a ~100× improvement. Eliminated schema-evolution table locks from 1000s+ to near zero.

      4–5s → 50ms (~100×)
      1,000s+ → near-zero lock time
      • MySQL
      • Java

    • Reliability engineering across 10+ services

      Established uptime targets, observability standards, alerting, and production-readiness practices across 10+ services and 100+ gRPC endpoints.

      10+ services
      100+ gRPC endpoints
      • gRPC
      • Grafana
      • Kusto/KQL

  2. Software Engineer Intern · LinkedIn

    Mountain View, CA

    May 2022 — Aug 2022

    • Built APIs and validation frameworks to enforce data-quality standards across LinkedIn's offline data lake, catching malformed assertions before they triggered production failures.

  3. Back-End Developer · Inspectinity

    India

    Aug 2020 — Dec 2020

    Safety- and quality-inspection SaaS for food and industrial companies — replacing paper and spreadsheet SOPs with a mobile-first audit tool.

    • Developed 20+ REST APIs in Python/Flask for high-volume backend workflows.
    • Optimized PostgreSQL queries for high-traffic components and built reusable auth, validation, and data-processing modules.

03 · The path

Seven cities.
One path.

From Patna to Toronto, each place shaped what I studied, what I built, and where I thought I could go next.

01Patna020304050607
3 countries7 cities11 years~21K km

01 · Origin

Patna

— Where it started

04 · Education

Where I learned.

ASU

M.S. in Computer Science · Arizona State University

Tempe, AZ · GPA 4.0 / 4.0

Jan 2021 — Dec 2022

Focused on distributed systems, large-scale data processing, and machine learning, with research in visual analytics and natural-language systems.

Publications

  • EuroVis 2023 · Computer Graphics Forum

    2023

    LINGO: Visually Debiasing Natural Language Instructions to Support Task Diversity

    Anjana Arunkumar, Shubham Sharma, Rakhi Agrawal, Sriram Chandrasekaran, Chris Bryan

    Developed a visual analytics system for identifying and reducing bias in natural-language task instructions, combining language-model outputs, semantic similarity, and interactive visualizations to uncover patterns across datasets.

Selected graduate work

  • Spring 2021 · Data Processing at Scale

    Geo-Spatial Hotspot Analysis

    Distributed Spark application performing spatial-temporal analysis on NYC yellow-cab pickup and drop-off data to surface statistically significant hotspots.

    • Implemented core spatial-query functions for range and distance queries.
    • Deployed on 4 EC2 servers — 14% reduction in execution time via distribution.
    • Apache Spark
    • SparkSQL
    • Scala
    • AWS EC2

  • Spring 2022 · Cloud Computing

    Auto-Scaling Distributed Inference Pipeline

    Distributed, fault-tolerant inference pipeline on AWS with an auto-scaler that provisions worker instances based on SQS queue depth. Demonstrated on a facial-recognition workload.

    • 50% reduction in total computation time via parallel workload distribution.
    • Elastic auto-scaling of workers (up to 20 EC2 instances) driven by queue depth.
    • AWS EC2
    • AWS SQS
    • AWS S3
    • Python

05 · Projects

What I build in the margins.

Built for myself first. Designed like they weren't.

Expenses — home overview with monthly spend and net worth

Designer & sole engineer

01 · 05/2026

Expenses

Personal finance across currencies, without another subscription.

A personal finance app for tracking expenses, accounts, net worth, and savings goals across currencies. Connects real accounts through Plaid and keeps both native and converted values using historical exchange rates.

  • Secure account linking — Plaid integration with AES-256-GCM at-rest token encryption and ES256-verified webhooks.
  • Multi-currency data model — Historical FX normalization via Bank of Canada rates, preserving both native and home-currency values.
  • Multi-user isolation — Workspace-scoped data layer with automated cross-user isolation tests in CI.
  • Next.js 15 (RSC)
  • React 19
  • TypeScript
  • Supabase (Postgres)
  • Plaid
  • Tailwind
GymPal — Today view with weekly calendar and workout card

Designer & sole engineer

02 · 06/2026

GymPal

A local-first workout tracker built around how I actually train. No accounts, no cloud, no subscriptions.

A native SwiftUI + SwiftData app for planning weekly workouts, logging sets, reps, and weight, and tracking progress over time. Designed for one-thumb use between sets — built to disappear into the workflow.

  • Zero dependencies — Pure native, offline-first. No accounts, no backend.
  • One-thumb operation — Gym-friendly controls with haptic feedback and reduced-motion support.
  • Built-in fitness features — Rest timer, MET-based calorie estimation, and weekly consistency tracking.
  • SwiftUI
  • SwiftData
  • iOS 26
  • Xcode
  • TestFlight

06 · Skills

The stack.

Languages, systems, and tools I reach for.

Languages

  • Java
  • C++
  • SQL
  • Python
  • Scala
  • JavaScript
  • TypeScript
  • Shell
  • Swift

Backend & Data Infra

  • Distributed Systems
  • Microservices
  • gRPC
  • REST
  • Protobuf
  • Kafka
  • Flink
  • Samza
  • MySQL
  • PostgreSQL
  • MongoDB
  • Couchbase
  • Elasticsearch
  • HDFS
  • Iceberg

Cloud & Tooling

  • AWS (EC2, S3, Lambda, DynamoDB, SQS)
  • Airflow
  • Docker
  • Kubernetes
  • Linux
  • DataHub
  • Grafana
  • Kusto/KQL
  • Vercel
  • Supabase

AI-native workflow

  • Claude Code
  • Codex
  • Cursor
  • Windsurf

07 · Contact

Let's talk about
a hard problem.

Toronto-based. Open to ambitious engineering problems and good conversations.

Email

rakhiagr2020@gmail.com

Resume

resume.pdf

LinkedIn

/rakhi-agr

GitHub

/rakhiagr

Instagram

/agrawal.rakhi