$ whoami

Arnav Gupta

C++ Systems & Backend Developer. Kernel IO at Pure Storage, ex-AWS, ex-Samsung Research — algorithmic optimisation, fast and clean C++, and low-level performance work.

Resume ↓ GitHub

1972

Codeforces · Candidate Master

23/101

ICPC Asia West Championship

Top 30

IICPC Codefest Global Regionals · of 39,000

3

SWE internships · 2× FTE offers

// 01. competitive programming

Proof of problem solving

ICPC — Asia West Continent Finalist

ranked 23rd of 101 teams

The ACM-ICPC is the world's oldest and most prestigious collegiate programming contest — over 50,000 students from 3,000+ universities across ~100 countries. Arnav's team "phirse lite lo" reached the Asia West Continent Championship (2026) after finishing 14th at ICPC India Regionals — a 2× Regionalist run.

certificate → regionals invitation → team on codeforces →

Codeforces — Candidate Master

Rating 1972 · top 1% globally

Multiple sub-200 contest ranks. Peak form on algorithms, math and speed.

profile →

JaneStreet Contests

BitShift 2026 — 6th of 2,500 · ICPC de Tryst 2026 — 7th of 3,000

Jane Street's own puzzle and programming contests — the ones quant firms watch.

IICPC Codefest — Global Regionalist

Top 30 in India · 139th of 39,000 overall

Ranked top 30 nationwide on the CS Fundamentals paper — Computer Architecture, Operating Systems, and Modern C++ (RAII, move semantics, templates, memory model).

certificate →

Practice Sheets

TLE Eliminators — ranked #1 · Striver A2Z — 420/420

First on the TLE Eliminators competitive programming sheet; every Striver sheet problem solved.

// 02. experience

Where I've shipped

Pure Storage current

Jun 2026 – Present

Member of Technical Staff 2 — Kernel IO

C++ · Linux

  • ▸Optimizing low-level kernel I/O paths to maximize data read/write throughput and minimize driver-level latency.
  • ▸Designing micro-benchmarking and testbenching suites to profile system execution and isolate hot-path bottlenecks.
  • ▸Implementing compile-time optimizations and multithreaded concurrency solutions in user-space C++.

Amazon — AWS

Jan 2026 – Jun 2026 · FTE offered

Software Development Engineer Intern

Java · Lambda · SQS · S3 · DynamoDB · Step Functions

  • ▸Architected scalable event-driven microservices powering the AWS Data Center Racks supply chain.
  • ▸Built an end-to-end integration and load-testing framework for high-throughput data APIs — improving Beta reliability from 0% to 97%.
  • ▸Shipped a notification routing system with zero-downtime rollout, cutting redundant compliance alerts by 90% across 377 global routes and 14 vendors.

Samsung Research Institute

May 2025 – Jul 2025 · FTE offered

R&D Intern — LLMs & Agentic Systems

Python · LangChain · RAG

  • ▸Built on-device Agentic AI and RAG pipelines for real-time inference over Digital Health datasets.
  • ▸Developed self-debugging SQL and Python agents with advanced visualizations.
  • ▸Won 1st place among 70 teams at the institute-wide Annual Best-Project Competition.

// 03. projects

Things I've built

High-Frequency Matching Engine

C++23 · Low-latency

Sub-microsecond limit order book: flat price arrays + intrusive lists for L1/L2 cache hits, zero-allocation hot path with arena pools and lock-free SPSC rings. SPSC ping-pong latency cut 231ns → 121ns via cache-line padding; tuned with Google Benchmark + Valgrind.

source →

Understanding Virality

Research · VLMs

A rubric-based multimodal framework predicting short-form edutainment engagement using VLM-extracted features — importance-weighted k-means + XGBoost, achieving Spearman ρ = 0.74 with real-world engagement. Published on arXiv, under review at ACL.

source →

CoCo Compiler

C · Compilers

DFA-based lexical analyzer and LL(1) predictive parser built from scratch — FIRST/FOLLOW set computation, parse-tree construction, validated across a range of test scenarios.

Audio Q-Former

PyTorch · Multimodal

Bridging SeamlessM4T audio representations to LLMs via a Q-Former architecture and contrastive alignment learning — connecting speech encoders to language-model reasoning.

// 04. skills

Tools of the trade

Languages C++20/23PythonJava 17C
Systems Lock-free concurrencyMultithreadingLinuxCMakeGoogle BenchmarkValgrind
ML / LLMs PyTorchLangChain / LangGraphVLMsXGBoostOpenCVNumPy / Pandas
Cloud AWS LambdaStep FunctionsDynamoDBS3SQSCDK

// 05. education & research

BITS Pilani

2022 – 2026

B.E. Computer Science, with Practice School — First Division

Coursework: OS, Compilers, Computer Architecture, Networks, DBMS, ML, Deep Learning, NLP, LLMs, Probability & Statistics, Combinatorics.

Understanding Virality

arXiv:2512.21402 · under review at ACL

A rubric-based VLM framework for short-form edutainment evaluation — first-author work with BITS Pilani + GenimeLabs.

read the paper →