Atharv Rawal — Backend & Systems Engineer

// backend · systems · cloud

Backend engineer

currently working on:

4thYear CS student, PES University · Bangalore

TMLRCo-authored survey on hybrid LLM inference

01. Experience

Summer
2025

Systems Developer @ PESU Innovation Lab

Project: PortMux — an async port multiplexer

Routed multiple protocols through a single exposed port via non-blocking I/O in Rust (Tokio).
First-packet inspection + YAML-driven routing, with custom Python protocol handlers.
Held ~8% latency overhead at 10,000 concurrent connections under Locust load testing.
Shipped as a systemd service, published to the AUR.

RustTokioPythonLinux

02. Projects

C++ / CUDA

GpuDedupe

Offloads SHA-256 hashing to the GPU while the CPU handles content-defined chunking — ~5x throughput over CPU-only dedup with byte-perfect restoration.

CUDAFastCDCLock-Free Ring Buffers

↗ Rust

Docksmith

Docker-alternative container runtime built from raw Linux primitives — no Docker, no runc, no containerd.

Linux NamespacesContent-Addressed StorageBuild Caching

↗ Rust

P2P File Sharing

Dual-stack P2P transfer over raw TCP, NAT traversal via an AWS signaling server, TLS end-to-end.

TCP/IPAWSTLS

↗ C++

ScreenStream

Turns a Windows monitor into a wireless second display — D3D11 capture, hardware H.264, delivered over WebRTC to any browser on the LAN.

D3D11 CaptureWebRTCMedia Foundation

↗ Rust

wifi-tui

Keyboard-driven TUI WiFi manager for Linux, fully async under the hood.

ratatuinmcli

↗ + More repos

03. Research

A Survey of Hybrid Inference Systems for Large Language Models

TMLR, 2026 · Co-author

Speculative decoding, model routing, and hardware-aware inference — organized into a four-level taxonomy that names the "Orchestration Gap" between inference latency and reasoning quality.

Read on OpenReview ↗

04. Contact

atharvrawal124@gmail.com

Open to backend, systems, and cloud roles & internships.