Aug 2023 – May 2027
Virginia Tech
Bachelor of Science in Computer Science, with minors in Mathematics and in Finance
GPA: 3.67 / 4.0
Portfolio
Varun Budati
Computer Science researcher and engineer applying reinforcement learning, optimal execution theory, and quantitative methods to financial markets.
Note graph
Every note, and every link between them. Click a node to read it, drag to rearrange. Prefer a list?
- Bio
- Research
- Projects
- Concepts
The graph couldn’t load. Browse the notes as a list instead.
The graph needs a wider screen. Browse the notes as a list.
About
Working at the intersection of research and implementation.
I work on how large orders get executed, and on what it costs when the act of trading moves the price you are trading against. Most of my time goes to the Dataism Lab at Virginia Tech, where we train reinforcement learning agents to liquidate a position and then try to break our own results before a referee does. The part I find most useful is not the model. It is the evaluation: deciding what would have to be true for a result to be real, and then building the control that could kill it.
Away from the lab I build things mostly to understand them. An options pricer to learn where the Greeks come from. A Unix shell in C to find out what a shell actually does. A pricing service for prediction markets, largely to see whether I could measure an edge honestly rather than just count winning bets. I am applying to PhD programs in operations research and financial engineering, and I am interested in how frictions at the level of a single order aggregate into risk at the level of a market.
Research Interests
Reinforcement Learning for Sequential Decision-Making
Optimal Execution & Market Microstructure
Statistical & Quantitative Methods in Finance
About me
- FocusML + Financial Engineering
- SchoolVirginia Tech
- DegreeB.S. Computer Science
- MinorMathematics & Finance
Education
Academic background
Jan 2010 – May 2023
IES Bhavans, Kuwait
Senior School Certificate Examination in Computer Science
Secondary School Examination in Information Technology · GPA: 94.3/100
Experience
Research, leadership, and competition results
IMC Trading Competition: 15th in the US (Top 0.12%)
Apr 2025- Placed 15th in the US out of 12,600 teams in IMC's Prosperity 3 trading competition.
- Optimized trade execution under latency constraints by engineering fair value estimators and implementing dynamic bid shading.
- Designed signal-driven market-making strategies by analyzing mean-reversion patterns and synthetic mispricing.
Undergraduate Commodities Competition (Houston, TX)
Oct 2025- Built a quantitative pricing engine in Python (NumPy, Pandas, SciPy) to back a multi-leg trade thesis (Short Jan '26 Waha Basis), calibrating a modified Ornstein-Uhlenbeck process to model structural temperature mean-reversion (κ = 0.2905) and local price drivers (R² = 0.40).
- Programmed a time-varying seasonal volatility matrix using an integrated cubic B-spline smoothing model, running 10,000 Monte Carlo path simulations to price winter Heating Degree Day (HDD) exotic call options for weather-risk hedging.
Quantitative Researcher, Dataism Lab for Quantitative Finance
Oct 2024 – Present- Researching optimal order execution by analyzing the market microstructure of Bitcoin trade data to develop and implement advanced trading strategies.
- Constructed benchmark execution algorithms in Python, including VWAP and TWAP, to analyze the market impact and transaction costs of trading Bitcoin.
- Engineered a Double Deep Q-Learning (DDQL) architecture in PyTorch, extended into a Mixture of Experts that routes limit order book episodes across K expert networks via an unsupervised K-Means router.
- Designed a repeated-training evaluation protocol (100 seed-paired runs per configuration, with parameter-matched dense controls) to separate real architectural effects from training-seed noise.
- Built the verification suite that audited the pipeline end to end, validating the execution model numerically against an independent re-implementation, reconciling campaigns across devices, and overturning several of the study's own published explanations.
Market Research Analyst, FinTech Club
Oct 2024 – May 2025- Led a project under Dr. Daniel Rodriguez replicating Evans & Archer (1968) via Python simulation, analyzing portfolio diversification and risk reduction.
- Utilized Pandas, NumPy, and SciPy on historical stock data to model risk vs. portfolio size, quantifying diversification benefits.
- Demonstrated and presented findings confirming that most unsystematic risk is mitigated with 10-20 assets.
Summer Researcher, Virginia Tech MAOP Program
May 2025 – Jul 2025- Conducted a mixed-methods research study on pedagogical patterns in introductory Computer Science to identify opportunities for improving student retention and engagement.
- Developed a quantitative framework with eight distinct categories through an open coding process to classify the thematic focus of programming examples.
- Utilized computational text analysis software (LIWC-22) and large language models to quantitatively assess the linguistic and contextual nature of course materials.
- Synthesized findings to reveal a high prevalence of abstract mathematical examples, providing data-driven recommendations for incorporating more culturally relevant and real-world applications in curricula.
Research Mentor, REACH Lab
Jan 2025 – Jun 2025- Guided undergraduate researchers within the Reimagining Equity and Accessibility in Computing for Humanity (REACH) Lab at Virginia Tech.
- Mentored peers on literature review best practices, database search strategies, and synthesizing findings for equitable computing research.
Undergraduate Research Assistant, REACH Lab
Mar 2024 – Jun 2025- Conducted an extensive literature review under the leadership of Dr. Ihudiya Finda Williams on Rural Computer Science Education.
- Synthesized over 80 articles to support a $500,000 National Science Foundation grant proposal.
- Trained six research assistants on database search methodologies, prompt design for scholarly queries, and rigorous literature review techniques.
Leadership
Leadership roles
Co-President, FinTech Club
May 2026 – Present- Attended Hokies on Wall Street twice, the club's trip to New York City, visiting firms including Elequin Capital, Evercore, and Citi.
- Organize and host speaker events with industry professionals from leading finance and technology firms to create career development and networking opportunities.
Treasurer, FinTech Club
Aug 2025 – May 2026- Managed the club's budget and all financial operations to fund initiatives and workshops for a community of 100+ members.
Student Summer Ambassador, MAOP Program
Fall 2026- Represented the MAOP program and helped support student engagement throughout the program.
- Gave multiple talks and speaker events to prospective students about the program and spoke about multiple research opportunities at Virginia Tech.
Publications
Research publications
SIGCSE TS 2026 · 2026
Bridging the Code Gap: Understanding Pedagogical Patterns in CS1 Programming Courses
Naomi Chioma Udenze, Varun Budati, Ihudiya Finda Ogbonnaya-Ogburu
An exploratory pilot study analyzing examples and exercises used in CS1 programming courses at top universities, investigating pedagogical patterns and diversity in computer science education.
Read publicationUnder Review · 2026
Mixture-of-Experts for Cryptocurrency Order Execution: Training Stability, Tail Risk, and Failure Modes
Ali Habibnia, Alexander Ardaiz, Varun Budati
A diagnostic study of whether mixture-of-experts (MoE) architectures actually help reinforcement-learning agents liquidate a large position in cryptocurrency markets. We retrain every configuration 100 independent times on Binance BTC/USDT limit order book replay: vanilla Double Deep Q-Learning, MoE at K in {2, 4, 8}, and dense networks parameter-matched to each expert budget. That way the comparison is between distributions of training outcomes rather than between single lucky runs. No configuration improves mean execution cost, and the benefits the architecture appears to have turn out to be separable. Across-seed training stability and within-policy tail risk are distinct axes, and they move in opposite directions. Most of the work went into building controls that could overturn our own result, and into reporting the ones that did.
Projects
My work
Optimal Trading Execution System with Reinforcement Learning
Implementation of the Almgren-Chriss model for optimal trade execution, enhanced with a reinforcement learning agent to minimize market impact.
Python
Reinforcement Learning
Quantitative Finance
View project
Options Pricing Calculator & Graph Using the Black-Scholes Model
A tool for pricing European options using the Black-Scholes model, with visualizations for the Greeks.
Python
Quantitative Finance
Streamlit
View project
MACD Trading Strategy Visualizer
An interactive web application to visualize the MACD indicator and backtest trading strategies on historical stock data.
Python
Streamlit
Technical Analysis
View project
Sports Betting Algorithm & Analytics System
A sophisticated analytics system to identify profitable betting opportunities using statistical models and machine learning.
Python
Data Analysis
Streamlit
View project
ShadeFinder
An urban shadow mapping project that helps identify shaded walking paths and cooler routes using geospatial analysis.
Python
Geospatial
Streamlit
SafeNight
SafeNight is a proactive safety ecosystem designed to prevent emergencies from escalating. By empowering users with real-time tools to monitor alcohol intake, stay tethered to a trusted community network, and access instant intervention, we are moving the needle from notification to prevention.
Hackathon
Full Stack
Collaboration
View project
minibash, a Unix shell in C
A shell written from scratch in C: its own lexer and recursive-descent parser for the shell grammar, job control, POSIX signal handling, and hand-rolled linked-list and hash-table primitives across 41 source files.
C
Systems
Parsing
Skills
Core tools and methods
Languages
Python
C
Java
JavaScript
SQL
R
Quant & ML
Machine Learning
Reinforcement Learning
Double Deep Q-Learning
Mixture of Experts
Financial Modeling
Statistical Analysis
Tools & Libraries
PyTorch
NumPy
Pandas
SciPy
Scikit-learn
Streamlit
Research
Research Writing
Notes
Things I'm working through
24 interlinked notes on market microstructure, execution, and the research itself. They're where I write out what I actually understand, including the parts I got wrong. Start anywhere.
Research
- CS Education Research What kinds of examples do intro programming courses actually use? SIGCSE TS 2026.
- Market Making and the IMC Competition 15th in the US out of 12,600 teams, and what trading under latency teaches you.
- MoE for Order Execution 100 retrainings of a mixture-of-experts crypto execution agent. It didn't win, and finding out why was the research.
- Replicating Evans & Archer (1968) How many stocks does it actually take to diversify? A simulation study.
Concepts
- Crypto Market Microstructure Why execution research built for equities doesn't transfer cleanly.
- Implementation Shortfall The gap between the price you decided to trade at and the price you actually got.
- Market Impact Why large orders cost more than small ones, in two distinct ways.
- Mixture of Experts Many specialists plus a router, instead of one generalist.
- Optimal Execution How do you sell a large position without moving the price against yourself?
- Reinforcement Learning for Trading Learning an execution policy instead of deriving one from assumptions.
- Seed Variance in Deep RL A single training run is a draw from a distribution, not a measurement.
- Specification Artifacts When your finding turns out to be a property of your config file rather than the world.
Contact
Get in touch
If you'd like to talk about research, finance, or collaboration, here are the best ways to reach me!