I'm Arya Chhabra.

About

The constant fiddling is what I love about software engineering. Finding the ugly edge case, the assumption everyone else took for granted, the hack that makes it click.

That's carried straight into how I build today — optimizing beneath the libraries everyone else simply plugs and plays and having the agency to take things from 0 to 1.

I pick things up fast and go deep once something grabs me — and I want that depth to help someone besides me. Freshman year that was picking up on running and learning to swim from scratch; now I'm running the Illinois Half-Marathon and pacing others through Illini Run Club. This year it's biking across America with our Bike for America team, raising money for cancer research.

Experience

Balancing high scale and low latency in distributed systems

Backend Engineer

UIUC FinTech Lab · Presented at 2026 STAC Research Summit, Chicago

Distributed systemsLow-latency infrastructureNetworking

2000×less cross-region sync error2 ms → 1 µs

16%lower latencyend to end

3hrs/wksaved weeklyvia automation

Things That Broke

System crashed in production

Problem

Data came in faster than the system could process and save it, so it backed up and crashed.

Fix

Added a buffer to hold incoming data and processed it in controlled batches instead of one piece at a time, keeping ingestion stable under full load.

WebSocket/TLS pipeline failures

Problem

Standard networking tools added tiny, unpredictable delays — bad news when you're measuring microseconds.

Fix

Bypassed those tools and timestamped data directly at the hardware level, using raw socket options for precise, consistent timing.

Creating agents that are reliable in executing large, complex enterprise workflows

AI Systems and Full-Stack Engineer

GEP

Agentic WorkflowsTesting and EvaluationBrowser and API Automation

10%less sourcing workload3,000-person procurement team at Chevron

30%less testing time20 developers

Things That Broke

Agent behavior was extremely hard to test

Problem

An AI agent's output could look correct on paper while still breaking what a user actually saw on screen — normal automated tests couldn't catch that.

Fix

Built a testing pipeline that checked the backend response and the real browser screen together, catching failures manual testing missed.

AI Engineer

Livegage AI

LLM systemsRetrievalAgentic workflowsDeterministic parsingBackend infrastructure

48% improved retrieval accuracy

54×faster graph editing

Things That Broke

Traditional RAG failed on documents

Problem

Standard AI search tools read documents as plain text, so they broke on regulatory files full of charts, diagrams and tables.

Fix

Built a retrieval system that reads documents visually — like a person looking at a page — instead of just their text, recovering the meaning that was being lost.