♣  THE HOUSE OF  ♦

ARNAV KASAT

♥  SYSTEMS & PROBABILITY  ♠

Computer Science & Mathematics at Purdue. I build things where correctness is falsifiable — storage engines you can crash, ledgers that can't lose money, and models that admit their own error bars.

Enter the casino → Read the resume
01

The Dealer

I'm an undergraduate at Purdue University, studying Computer Science and Mathematics with a minor in Finance.

My work keeps landing in the same place: systems where a wrong answer is expensive, and where you can prove you didn't give one. A key-value store that survives a thousand SIGKILLs without losing an acknowledged write. A payment ledger that stays zero-sum through 500 injected failures. A matching engine held to a 9µs p99.

The other half is probability — market regimes, stochastic volatility, reinforcement learning on portfolios. Which is also why this site is a casino: it's the most honest place to show that I know where the edge sits, and who it belongs to.

Coursework

Design & Analysis of Algorithms Data Structures & Algorithms Systems Programming Computer Architecture Object Oriented Programming Discrete Math Probability Linear Algebra Introduction to AI Machine Learning

Off the clock

Quantitative finance, algorithmic trading, probability theory, poker, soccer, HYROX, and piano.

02

Experience

Software Engineering Intern — AI Innovation Program

Capital One · Cloud Operations Resilience Engineering

May 2026 – Aug 2026 · McLean, VA

  • Built an MCP server on ECS Fargate for 500+ users to query petabyte-scale telemetry, cutting triage time by 92%
  • Engineered a Spark Connect pipeline over an 889B-row Iceberg lake on EMR, routing multi-TB/day workloads
  • Enforced tenant isolation and blocked unauthorized LLM data access using RBAC and AWS STS credentials
  • Protected EMR with sub-100ms scan checks, partition filters, and query limits; shipped to production via CI/CD

Undergraduate Researcher — Machine Learning & Quantitative Finance

Purdue University

Jan 2024 – Present · West Lafayette, IN

  • Developed an autoencoder-based coordinate system for S&P 500 returns to identify latent market regimes
  • Modeled regime transitions using Dynamic Mode Decomposition and sparse latent equations through SINDy
  • Built a PyTorch, PyDMD, and PySINDy pipeline and authored a research paper on regime-based market modeling

Software Engineering Intern

AlphaGeo

May 2025 – Aug 2025 · New York City, NY

  • Built a Python pipeline to unify 100K+ BLM, USFS, and legislative land records using PostGIS, Pandas, and Shapely
  • Automated ingestion via REST APIs and LLM-based geocoding to standardize incomplete spatial records
  • Created dashboards and U.S. risk heat maps (Streamlit, PostGIS, Plotly) to surface high-privatization parcels

Software Engineering Intern

PI Semiconductor

May 2024 – Jul 2024 · Dallas, TX

  • Built an internal Python tool with an Excel-style UI (Tkinter, openpyxl) to streamline analog chip spec input
  • Structured and validated spec data using Pydantic and NumPy to generate Verilog-compatible simulation inputs
  • Automated ingestion into hardware validation flows, reducing simulation turnaround time by 4.6%

Project Manager — Algorithmic Trading Team

ML@Purdue

Aug 2024 – Apr 2025 · West Lafayette, IN

  • Led a 6-member team developing ETF arbitrage strategies using Heston stochastic volatility modeling

Resident Assistant

Purdue University

Aug 2024 – Present · West Lafayette, IN

  • Primary contact for dorm issues, mentor to 100+ residents, and organizer of community events
03

Projects

Crash-Resilient LSM-Tree Storage Engine

Rust · Linux · Criterion

  • Embedded ordered key-value store written from the ground up: WAL, MemTables, SSTables, tombstones, atomically-swapped manifests
  • Bloom filters, sparse indexes, leveled compaction, and an LRU block cache cut read latency by 64%
  • Sustained 95K writes/sec and 62K reads/sec over 20GB with zero committed-write loss across 1,000 crash tests
  • No external crates in the library — CRC-32C, Bloom filter, LRU, block format, varints and PRNG all hand-rolled
Source →

Idempotent Payment Engine

Go · PostgreSQL · Kafka

  • Concurrent Go payment engine on a double-entry PostgreSQL ledger, sustaining 2.1K TPS at 58ms p95
  • Idempotency keys, a transactional outbox, and Kafka consumers deduplicated 1M retries — exactly-once effects over at-least-once delivery
  • Held zero-sum ledger invariants across 500 injected failures, recovering in under 8 seconds
  • Ships an operator dashboard that arms failure injection at runtime and recomputes every invariant once a second
Source →

Hyrox Tracker

Next.js · TypeScript · Supabase · Tailwind

  • Training tracker for HYROX doubles — a 20-week plan generated from a race date, daily check-ins, and set-by-set logging
  • Apple Health ingest plus a weekly review that proposes plan changes you accept or reject
  • Row-level security on every table, magic-link auth, and an installable iOS PWA
Source →

Matching Engine & Limit Order Book

C++17 · CMake · GoogleTest · Linux

  • Matching engine with price-time priority, partial fills, cancellations, and order amendments
  • O(1) order lookup and O(log P) price-level updates via hash maps and FIFO price queues
  • Processed 750K orders/sec at 9µs p99 latency across 10M randomized order events

Reinforcement Learning Portfolio Optimization

PyTorch · QuantConnect

  • Deep Q-Learning framework for dynamic asset allocation with adaptive portfolio weights
  • Tuned hyperparameters and risk-adjusted reward functions, reducing drawdowns by 15.4%
  • Validated on QuantConnect against real-time market data for a 23.7% higher Sharpe ratio

Latent Market Regimes via DMD & SINDy

PyTorch · PyDMD · PySINDy

  • Autoencoder coordinate system over S&P 500 returns to surface latent market regimes
  • Regime transitions modeled with Dynamic Mode Decomposition and sparse governing equations
  • Research paper authored on regime-based market modeling

Stochastic Volatility & Dispersion Trading

Python · Quantitative Modelling

  • Dispersion strategy built on the Heston stochastic volatility model to capture option mispricing
  • Monte Carlo risk analysis plus an automated delta-neutral hedging system
  • Mispricing detection engine with broker API integration for execution-ready signals

BoilerMate

Python · CrewAI · ChromaDB · Streamlit

  • Purdue AI assistant doing retrieval over course, grade, exam, and university resource data
  • BeautifulSoup pipelines ingested 13K+ course records and indexed grade distributions in ChromaDB
  • Role-based CrewAI agents sharing context in Streamlit for planning and resource discovery
04

The Floor

Four games, and the math in all of them is real. The blackjack trainer grades you against published basic strategy and a true Hi-Lo count, and the roulette table shows exactly how much the house takes on every bet you place.

♠ Blackjack — Count Trainer

6-deck shoe · S17 · DAS · Hi-Lo

  • Play real hands, then call the count and name the play — both graded against the true count and the basic-strategy table
  • Running count, true count, decks remaining, and your accuracy on both drills, live
Sit down →

♦ Roulette — The House Edge, Visible

European & American wheels

  • Place any standard bet and see its exact payout, win probability, and expected value before you spin
  • Watch your realized return converge on −2.70% (or −5.26%) as the sample grows
Place a bet →

♥ Odds Oracle

Rapid-fire probability

  • Cards, dice, coins, birthdays, Monty Hall — estimate the probability, get scored on how close you were
  • Every answer comes with the derivation, not just the number
Take a guess →

♣ Pickleball — Retro Sim

Canvas · rally scoring to 11

  • CRT-green pickleball with a real kitchen line: volley inside it and you lose the rally
  • Mouse, touch, or arrow keys. Three difficulty settings, best of one.
Serve →

Hall of Fame

Best results anyone has posted, across every visitor. Roulette has no board — there is no skill in it to rank.

Loading…

05

Technical Skills

Languages

PythonC / C++Rust GoJavaTypeScript JavaScriptSQLScala CUDARBash HTML / CSS

Cloud & Infrastructure

AWS ECS / FargateEMRSTS CloudWatchGCP Cloud RunCloud SQL DockerLinuxCI/CD GitHub Actions

Data & AI Systems

SparkIcebergKafka PostgreSQLPostGISMCP FastAPIPydanticPyTorch ChromaDBTensorFlowScikit-learn XGBoost

Frameworks & Libraries

ReactNext.jsNode.js FlaskStreamlitSupabase PandasNumPyMatplotlib PlotlyShapelyGit
06

Cash Out

Open to software engineering and quantitative internships. The fastest way to reach me is email.

Email LinkedIn GitHub