I build the infrastructure other engineers stop having to think about.
Backend engineer at PayPal, on the platform that sells buy-now-pay-later and credit card loans to institutional investors. Before that, two years at Pure Storage building durable workflow systems, and an MS in Computer Science from NC State. Lately, agents.
Most recent first. Scroll to rewind the trip; click any bar to jump.
Full-timeJan 2026 to now / Scottsdale
PayPal
Software Engineer
01Backend for PayPal's receivables sales to institutional investors: hundreds of millions every week, through forward-flow and securitization programs of up to EUR 65B.
02Moved loan selection from Java into BigQuery SQL: the weekly batch job went from about an hour to 45 seconds, verified side by side against the old job; then designed the selection logic for a new credit card program.
03Built an in-process test framework for Spring Batch jobs: 1,000+ scenarios in about a minute, cutting CI from 2-3 hours to 30 minutes.
04Built a Claude Code multi-agent system used by a few teammates: bug-finding reviewers checked by a critic agent, and a support agent that investigates failed runs and proposes a tested fix for approval.
More
05Own alerting and reconciliation for loan sales across 50+ live alert scripts, consolidating them into one config-driven framework.
06Self-initiated an internal analytics hub with a text-to-SQL LLM assistant; its SQL gateway was adopted by a senior staff engineer.
BigQuerySpring BatchSpring BootJava
PHXBefore this / 1,890 mi / flew Dec 2025 / after graduatingRDU
EducationAug 2024 to Dec 2025 / Raleigh
NC State
Master of Computer Science
01Built virtual memory for the XINU operating system kernel in C and x86 assembly: demand paging, two-level page tables, an inverted page table and clock eviction. Wrote the test harness that exposed an off-by-five eviction bug, and traced it to unaccounted reserved frames. Scored 100/100.
02Wrote a C++ ns-3 simulation of datacenter congestion control (DCTCP vs BIC), and built a reproducible benchmark across four VMs comparing HTTP/1.1, HTTP/2 and BitTorrent throughput from 10 kB to 10 MB.
03Open source: built a review grading dashboard for Expertiza, NC State's peer-review platform. The React + TypeScript front end is merged; the Rails API and schema migration are in review.
044.0/4.0 GPA, with A+ in Operating Systems, Computer Networks and Algorithms.
CC++TypeScriptRuby on Rails
RDUBefore this / 1,890 mi / flew Aug 2025 / back to NC StatePHX
During NC State
InternshipMay 2025 to Aug 2025 / Scottsdale
PayPal
Software Engineer Intern, Backend
01Built a complete Spring Batch pipeline syncing 6.5M accounts into BigQuery to tackle a data-mismatch backlog costing $75K a month; the team adopted its design for production.
02Cut about 90% of manual SQL fixes with idempotent, multi-threaded processing, bounded retries and an audit table for every failed record.
03Built a lightweight JUnit and Mockito test framework covering every batch job component.
BigQuerySpring BatchJavaJUnitMockitoTesting FrameworkBNPL Systems
PHXBefore this / 1,890 mi / flew May 2025RDU
NC State begins here / Aug 2024
RDUBefore this / 8,730 mi / flew Aug 2024BLR
Full-timeJul 2022 to Jul 2024 / Bengaluru
Pure Storage
Member of Technical Staff 3
01Designed and led Auto Sim Rehab, an event-driven Temporal, RabbitMQ and Redis system that repairs failed testbeds automatically, cutting MTTR from over 5 hours to 10 minutes across 2,000+ testbeds. Drove promotion to MTS-3.
02Led the release gate for every VM image: each build is booted, health-checked and smoke-tested before it can ship. Demoed to the CTO.
03Built a desired-state inventory system on Nautobot that discovers every switch and host and flags drift, raising accuracy from about 60% to 99%.
04Led the simulator image upgrade from Ubuntu 14 to 22, cutting deployment time 90% with parallel batch pipelines.
More
05Mentored the intern who took over Auto Sim Rehab; with one teammate, kept the shared test framework running after its team was laid off.
01Built CLI2API, turning switch CLI commands into REST calls and cutting data fetch from 135 seconds to 15; demoed to a VP.
02Cut regression triage from about an hour to five minutes with rule-based auto-triage of weekly Jenkins runs, adopted by the team.
03Proposed and built a Perl-to-Python transpiler to migrate the legacy test suite instead of rewriting it by hand.
REST APIsNetwork TestingCLI ToolsPLYPerlPython
EducationJul 2018 to May 2022 / Bengaluru
PES University
B.Tech in Computer Science and Engineering
01Core Computer Science: Data Structures, Design and Analysis of Algorithms, Operating Systems, Database Management Systems, Compiler Design, Object-Oriented Analysis and Design with Software Engineering
02Networks & Security: Computer Networks, Computer Network Security, Information Security, Web Security, Network Applications
03Data Science & AI: Introduction to Data Science, Machine Intelligence, Big Data, Internet of Things (IoT), Cloud Computing
04Software Engineering: Software Testing, Introduction to DevOps and Site Reliability
03 / Projects
Selected builds
Three things I built outside the day job, problem first.
01 / Go / gRPC / MySQL
KapLabs IAM
Every app we shipped needed its own logins, users and permissions. I built one multi-tenant identity platform they all share.
auth.kaplabs.dev
0:00 / 0:00
gRPC endpoints
45
authz test coverage
94.4%
lines of Go
~9,000
tenancy levels
4
How it is built
Decisions worth asking about
01
RS256, not HS256
Tenants can verify a token but never mint one; a shared HS256 secret would give every app signing power over every other tenant.
02
One column, two worlds
An app's identity scope is either isolated users or the org's shared pool, so the same code gives isolation or single sign-on.
03
A gateway that cannot drift
The HTTP/JSON API is reflected from the gRPC service table and runs the same handlers and interceptors, so the two surfaces never disagree.
Our meal plan, pantry and shopping list never agreed, so food expired and the store run was guesswork. Platr keeps them on one truth: it cooks from what expires first and writes the shopping run.
mealpal.kaplabs.dev
0:00 / 0:00
MCP tools for Claude
7
test files
56
database tables
22
tabs: Today, Plan, Pantry, Shop
4
Decisions worth asking about
01
First expired, first out
Cooking draws stock from the lot that expires soonest, so the pantry spends what would otherwise be thrown away.
02
Ranges, not guesses
Pantry run-out is predicted from real consumption: mean daily use plus its variance gives a date range, collapsing to a point when history is too thin.
03
Read-only by construction
The MCP server imports no mutating function, so Claude can read the whole household's data with no write path to guard.
PayPal / Scottsdale, AZ
Pure Storage
NC State MS
