Austin, TX · CS + Mathematics · UT Austin ’28

Aryan Kondapally

I’m a computer science and math double major at UT Austin. I like building things end to end and understanding them all the way down — lately that’s been LLM evaluation infrastructure at Uber, compliance automation at IBM, and my own research on graphs, markets, and quantum hardware.

OpenSummer 2027 internships — SWE · quantitative development · ML systems
NowAI Systems Engineer (Contract) · Uber — LLM agent evaluation infrastructure
2026Software Engineer Intern · IBM — compliance & deployment-readiness automation
PrevQuantum computing research · UT Dallas — published co-author
Email GitHub LinkedIn Resume (PDF)
Aryan Kondapally
−50%
Manual application review at IBM, cut by Beacon — the compliance framework I built there
+1.7 pts
Cora accuracy gain from letting a GNN learn its own geometry — up to +7.0 pts on synthetic benchmarks
3 papers
Quantum hardware crosstalk (published co-author) · price-reaction prediction · graph geometry

Experience

May — Aug 2026

IBM

Software Engineer Intern

  • Built Beacon, an internal Java/Gradle framework that automates compliance and deployment-readiness validation for Docker, Kubernetes, and OpenShift applications.
  • Implemented extensible log, security, and configuration validators with automated compliance scoring and reporting — cutting manual application review by 50%.
  • Developed Docker and file-based collectors and configurable CI/CD validation profiles used across IBM’s webMethods and API Connect teams.
Sep 2025 — Present

Uber

AI Systems Engineer (Contract)

  • Build deterministic evaluation packages for LLM agents — task specifications paired with runnable commands, automated tests, and rubric-aligned expected outputs across 10+ tasks.
  • Developed pytest validation harnesses covering execution behavior, output integrity, tensor dimensions, and metric thresholds across environments.
  • Implemented discrepancy logging between agent-generated and reference solutions, enabling systematic triage of recurring model failures.
Jun — Sep 2023

The University of Texas at Dallas

Software Engineering Research Intern — Quantum Computing (while in high school)

  • Built Python/Qiskit pipelines to transpile, route, and benchmark quantum circuits across IBM Q qubit layouts and hardware topologies.
  • Identified an 80% simulation-to-hardware gap by analyzing routing, parallel execution, qubit proximity, noise coupling, and crosstalk.
  • Collaborated with PhD researchers and co-authored a published paper on CNOT-induced crosstalk and hardware connectivity.

Projects

CPU emulation, backends, smart contracts, and ML research. Everything linked is public on GitHub.

Learning Geometry on Fixed Topologies

ML Research · 2026

A framework that makes the geometry of a graph trainable: per-edge metric weights, kept positive-definite through softplus, are optimized jointly with the network on a fixed topology — so the model learns the space it computes in.

  • Four families of geometric regularizers: Ollivier–Ricci and Forman–Ricci curvature, Laplacian smoothness, volume penalties against metric collapse, and deviation controls.
  • Two-timescale optimization — separate AdamW groups for network vs. geometry parameters, with metric gradients clipped ~10× more aggressively to stabilize joint training.
  • Controlled three-way comparison at identical parameter budgets: fixed baseline, learnable metric, and a frozen-metric ablation isolating the effect of geometry itself.

PyTorch · PyTorch Geometric · NetworkX · pytest · LaTeX paper draft

Accuracy, mean of 5 seeded runs
BenchmarkFixedLearnedΔ
Two-Moons0.8420.912+7.0
Swiss Roll0.7610.823+6.2
Cora0.8140.831+1.7

Fair Value Gap Degree Analysis

Quant Research

Implementation of my self-published research paper: a RANSAC-derived “degree” metric, fit to 1-second tick data during gap formation, that predicts the strength of price reactions at Fair Value Gaps.

  • Multi-horizon reaction engine measuring boundary retests over 5/15/30/60-minute windows.
  • Hypothesis validation via two-sample t-tests and chi-square contingency tests, plus a strategy backtest.
  • Multi-source ingestion (yfinance, MT4, Binance) with multiprocessing-parallel per-gap analysis.

Python · pandas · scikit-learn (RANSAC) · SciPy · seaborn

Read the paper

chARM Emulator & Cache Simulator

Systems · C

A full-system chARM-v3 emulator with a five-stage pipelined CPU — fetch, decode, execute, memory, writeback — built in C for UT Austin’s computer architecture sequence.

  • Hazard detection and control with stall, flush, and forwarding logic to resolve pipeline hazards.
  • Trace-driven, configurable cache simulator integrated with the pipeline to model memory latency.
  • Performance evaluated across block sizes, associativities, and replacement policies.

C · computer architecture · pipelining · caching

HornsTraffic

Full-Stack iOS

Live gym-occupancy tracking for UT Austin’s Gregory Gym and Rec Center — an MVVM SwiftUI app over a JWT-secured Node.js/Express/MongoDB API, built end to end.

  • Server-side gap-filler synthesizes decayed occupancy points when updates stall — mirrored client-side so charts never break.
  • bcrypt-hashed admin auth with 12-hour JWTs guarding the write path.
  • Swift Charts with Catmull-Rom interpolation; Combine-driven 15-minute auto-refresh.

SwiftUI · Swift Charts · Node.js · Express · MongoDB · JWT

DeFi Staking Protocol

Smart Contracts

A three-contract Ethereum protocol: capped ERC-20 token, staking vault, and stake-weighted on-chain governance with a 2-day timelock — 99.4% line coverage, with a hardhat-deploy pipeline targeting Sepolia.

  • MasterChef-style O(1) reward accounting via a 1e18 fixed-point accumulator — no loops over users.
  • Full 7-state proposal lifecycle executing arbitrary calldata; voting power read live from stakes.
  • Ownable2Step, ReentrancyGuard, SafeERC20, checks-effects-interactions throughout.

Solidity 0.8.20 · Hardhat · OpenZeppelin v5 · TypeChain · Ethers v6

Austin Crime Ontology

Data Engineering

Ontology-driven geospatial analytics over 500,000 Austin PD incident records — six linked object types modeled in DuckDB so queries traverse relationships instead of scanning flat tables.

Python · DuckDB · GeoPandas · scikit-learn · H3 · Folium

  • Entity resolution for inconsistent addresses: H3 hex bucketing to avoid O(n²) comparisons, haversine matching within buckets, transitive merge-chain collapse.
  • DBSCAN hotspot detection (haversine metric on a ball tree) with census-tract demographic enrichment.
  • Ingestion hardened against live API schema drift, with Parquet caching and Folium map exports.
  • The README documents known coordinate-quality issues and what needs fixing before drawing conclusions from the data.

Research

2023Published · Co-author

The Impact of Limited Qubit Connectivity on CNOT Gates: Understanding and Mitigating Crosstalk Errors in Quantum Computation

Co-authored with UT Dallas PhD researchers. Benchmarked circuits across qubit layouts and topologies to characterize how routing, qubit proximity, noise coupling, and crosstalk drive the gap between simulated and on-hardware performance.

2025Self-published

Quantifying Fair Value Gaps: A Novel Metric for Price Reaction Prediction in Financial Markets

Proposes the “degree” metric — a robust RANSAC slope fit to tick-level price action during gap formation — and validates that low-degree gaps produce measurably stronger price reactions.

Paper Code
2026Working paper

Learning Geometry on Fixed Topologies for Graph Neural Networks

Frames GNN message passing as optimization on a learnable discrete Riemannian metric, with curvature, smoothness, and volume regularization — and characterizes exactly when geometry learning helps and how it fails.

Code & draft

Skills

Languages Java · Python · C · TypeScript · JavaScript · SQL · Swift · Bash
Frameworks PyTorch · React · Node.js · Express · SwiftUI · Qiskit · pytest
Infrastructure Docker · Kubernetes · Git · Linux · Gradle · MongoDB · OpenTelemetry
Coursework Data Structures · Algorithms & Complexity · Operating Systems · Computer Architecture · High-Performance Programming · Artificial Intelligence

Get in touch

I’m looking for Summer 2027 internships in software engineering, quantitative development, or ML systems. Email is the fastest way to reach me.

GitHub LinkedIn Resume (PDF)