Aarav Jamdar
Education
University of California, Berkeley
Expected May 2030B.A., Mathematics, Computer Science & Data Science
- College of Computing, Data Science, and Society
- GPA: 4.00
Leland High School
Aug 2022 – May 2026High School Diploma · San Jose, CA
- GPA: 4.39 · SAT: 1570
- 13 AP Exams — all 4s and 5s
- AP Calculus BC, AP Statistics, Multivariable Calculus (DE), Differential Equations (DE), Linear Algebra (DE)
Experience
University of California, Santa Barbara
Jun 2025 – Jul 2025Summer Research Academy — Santa Barbara, CA
- Designed 3D aerodynamic vehicle models inspired by falcon wing and nostril anatomy using computational modeling tools.
- Found a 14% improvement in airflow redirection efficiency for the falcon-inspired grille, and a 6% lower drag coefficient over industry baseline designs.
- Validated results against physical wind-tunnel prototypes.
- Co-authored a research paper and presented findings at a capstone seminar to 200+ attendees.
Cambridge University
Jun 2024 – Dec 2025Student Researcher — Remote
- Used Argonne National Lab's GREET model to compare life-cycle CO2 emissions of battery-electric and hydrogen fuel-cell vehicles under the mentorship of Dr. Jordi Ferrer.
- Designed computational simulations, collected and analyzed experimental data, and authored a full research manuscript.
- Paper accepted and published in the Journal of Emerging Investigators (JEI), a peer-reviewed STEM journal with a ~30% acceptance rate.
Distinctions
- 4th / 18,000 nationally in Public Forum Debate
- 1st at James Logan
- 3rd at the Cal Invitational
- 7th at NSDA Nationals
- 2× Gold bids at the Tournament of Champions
- Only 7-0 at Glenbrooks
- NSDA Academic All American
- Congressional Award Silver Medal
- SAT: 1570
- 3x CHARGE Award recipient
Projects
RL Trading Agent
DQN built from scratch, deployed to live paper trading with a real risk layer.Built a Deep Q-Network from scratch (Q-network, experience replay, Polyak-averaged target network), validated on CartPole-v1 before touching market data. Trained on SPY with a chronological train/validation split, then found via direct action-distribution inspection that the policy had converged to buy-once-and-hold, a direct result of rewarding raw dollar P&L on a steadily rising asset. Deployed the frozen policy for live paper trading through Alpaca with a drawdown kill switch and equity-capped position sizing, running unattended via GitHub Actions.
EquiFace v2
Facial recognition bias audit reproducing the Gender Shades disparity.Trained a ResNet18 gender classifier via transfer learning on FairFace (28,000 images, 7 races × 2 genders), reaching 88.09% accuracy on a 5,600-image held-out set. Audited the model for demographic bias using bootstrap-resampled confidence intervals, per-race TPR/FPR gap analysis, and a two-proportion z-test, surfacing a statistically significant disparity (z = 5.51, p ≈ 3.7×10⁻⁸), reproducing the Gender Shades finding. Built a custom stratified sampler to hit exact per-subgroup image quotas.
Statistical Arbitrage Backtester
Cointegration-tested pairs traded on a rolling z-score, bias-corrected.Tested candidate stock pairs for cointegration and derived hedge ratios via OLS regression to construct a market-neutral spread. Simulated long/short positions from a rolling z-score with look-ahead-bias correction, evaluated via Sharpe ratio, max drawdown, and win rate against the risk-free rate.
Core logic verified with pytest against hand-calculated values; methodology and limitations written up in the README rather than left implicit.
Publications
Transitioning to a Zero-Emission Future: A Life Cycle Comparison of Battery Electric & Fuel Cell Vehicles
2025Journal of Emerging Investigators (JEI)
Used Argonne National Lab's GREET life-cycle model to compare cradle-to-grave CO2 emissions of battery-electric and hydrogen fuel-cell vehicles, finding EVs currently emit less under the U.S. grid but HFCVs could overtake them under a fully renewable one.