Atharva Arbat

Systems engineer. I write the layers most people skip — schedulers, kernels, wire protocols.

• •AI agent infrastructure, distributed systems, and open protocols excite me.
• •I'd rather prove a system is fair than just claim it is — AgentKernel ships with a closed-form fairness bound.
• •Open to SWE and AI infrastructure internships — currently building AgentKernel and other multi-agent systems experiments.

Experiences

OVRT

AI Agent Engineer Intern

Jan 2026 - Jun 2026

Remote

6

LLM Agents

20%

Latency Cut

1,000s

Events / Day

Jan-Jun

Term

• •Built the backend of a Shopify automation platform in TypeScript: 6 LLM agents coordinated through BullMQ job queues, with LangFuse tracing and exponential-backoff retries over thousands of events daily
• •Designed a secure Hono API proxy with HMAC request signing between internal services and third-party AI APIs (OpenAI, Sarvam AI, Vercel AI SDK); cut service latency 20% with connection pooling
• •Ran a 24/7 monitoring agent in production for automated store analytics across merchant accounts

TypeScriptHonoBullMQLangFuseOpenAI APISarvam AIVercel AI SDKHMAC Auth

Vegapod Hyperloop

Navigation and Control Lead

Aug 2023 - Aug 2025

Pune, India

9 Engineers

Team

11ms

Latency

8th / 16

Rank

Zurich, Groningen

Competitions

• •Architected the pod's master control unit in C++: a multi-process, multi-threaded TCP/UART system that held 11 ms average end-to-end latency in live runs at European Hyperloop Week 2024 (Zurich) and 2025 (Groningen)
• •Built a Qt C++ desktop GUI and a React web interface for live telemetry display on Linux single-board computers
• •Designed a custom binary protocol over UART and TCP with heartbeat, CRC16 error detection, and auto-reconnect for fault-tolerant state sync between pod and base station
• •Led the 9-engineer navigation and control subsystem to an 8th place finish among 16 international teams; owned the technical roadmap and reviewed every change to the control stack

C++UARTTCP/IPCRC16QtReactEmbedded Systems

BytesWrite

Frontend Developer Intern

Aug 2024 - Nov 2024

Pune, India

150+

Employees

54 → 91

Lighthouse

4.2s → 2.5s

Load Time

35+

Components

• •Built an internal operations platform (Next.js, TypeScript) from scratch; adopted by 150+ employees within 3 months, replacing manual spreadsheet workflows
• •Raised the Lighthouse performance score from 54 to 91 with SSR, incremental static regeneration, and route-level code splitting, cutting initial load from 4.2s to 2.5s on a throttled 3G profile
• •Decomposed a monolithic UI into 35+ reusable TypeScript components, cutting code duplication by roughly 30%

Next.jsTypeScriptSSR / ISRComponent Architecture

View All

Projects

AgentKernel

Shipped

A Go scheduling kernel for multi-agent LLM systems with a proven fairness bound — 0.9998 Jain fairness index at 1,847 ns/call overhead.

Fair Scheduling

Multi-Agent Systems

View Project

ConcurrentKV

Shipped

A Redis-style key-value store over a custom binary TCP protocol, sustaining 49,000 ops/sec with a lock-free MPSC queue.

TCP Sockets

LRU Cache

View Project

ARBOS

Shipped

A bootable 32-bit x86 operating system built from scratch: multiboot bootloader, paging, interrupts, and device drivers.

x86 Assembly

Bootloader

View Project

Auth4Agent

Shipped

An open-source decentralized identity protocol for AI agents — Ed25519 keypairs and challenge-response auth where private keys never leave the agent.

Ed25519

View Project

View All

Skills & Technologies

Go

TypeScript

JavaScript

Python

C++

C

React

Next.js

Node

Hono

Express

FastAPI

LangChain

Vercel AI SDK

PostgreSQL

MongoDB

Prisma

JWT

Tailwind

shadcn

Docker

Linux

Git

Github

CMake