I build quantitative models end to end, then try to break them. My work spans the full stack of a quant team: a C++20 matching engine that clears ~21M messages a second, an execution model tested against a reinforcement-learning agent on 207,361 real bars, a portfolio optimizer that prices its own trading costs, and the VaR and FRTB numbers a regulator signs off on. Three years running market-risk infrastructure at Nomura taught me what production demands. The MS in Financial Engineering at NYU Tandon (3.84 GPA) gave me the theory to go further. I use point-in-time data, ship tests with every model, and report what failed alongside what worked.
I am not narrowing myself to one seat. Research, trading, development and risk draw on the same habits: careful data, honest statistics and code that is tested before it is trusted. Here is the evidence for each, with the résumé written for it.
I have tested cross-sectional alpha on 495 names with Newey–West inference and false-discovery control, trained purged walk-forward learners, and audited my own regime model against NBER dating, finding it too jumpy. I treat every t-statistic as a claim that has to survive its own robustness check.
I calibrated Almgren–Chriss to a live Deribit book and stress-tested it against a policy-gradient agent, fitted SABR to the BTC volatility smile, and am now building a Deribit options market maker with latency-aware fills, delta hedging and P&L after fees.
I wrote a zero-allocation C++20 matching engine that clears ~21M messages a second and an options library that prices three independent ways and agrees to within 0.6%. At Nomura I built Python and SQL pipelines that removed 50+ hours of manual work a month.
At Nomura I owned VaR and sensitivity models across equities, rates, FX and commodities, and Basel 2.5 and FRTB submissions reconciled 100% to front-office systems. I have since built a historical CVA that measures wrong-way risk by experiment rather than assuming it.
A C++20 engine that replays recorded Deribit order books, runs several quoting strategies side by side, simulates fills with latency, hedges delta on the BTC perpetual, and reports P&L after fees and adverse selection. The same engine then runs live on paper. If a strategy loses money after costs, that is the result.
Every figure below is reproducible from its repository, covered by tests in CI, and reported with the assumption that could break it. Where a published effect failed to replicate, I say so. That is the standard I would bring to any desk: a number you can defend in front of a risk committee, not one that only survives a slide.
I tested eight cross-sectional signals on 495 S&P 500 names over 2010–2026: six built from prices and two from the tone of 15,575 SEC filings. Each is scored by Spearman rank information coefficient with Newey–West t-statistics, then Benjamini–Hochberg across all eight tests. Only Amihud illiquidity survives. Low volatility and the lottery signals come out with the wrong sign.
So the obvious move, z-score everything and average, loses 7.8% a year at a Sharpe of −0.57. Learning the combination with LightGBM under purged walk-forward CV with a one-month embargo, charging 10 bps a side, reaches 0.79 out of sample: within a quarter-point of the best single signal at two-thirds of its drawdown.
| Signal | Mean IC | NW t | Hit | BH q |
|---|---|---|---|---|
| amihud | 0.032 | 5.32 | 62% | 0.000 |
| tone_chg_naive | 0.007 | 1.37 | 57% | 0.46 |
| mom_12_1 | 0.008 | 0.69 | 54% | 0.60 |
| strev_1m | 0.006 | 0.64 | 48% | 0.60 |
| tone_chg | 0.003 | 0.46 | 55% | 0.65 |
| skew_120d | −0.004 | −0.70 | 46% | 0.60 |
| max5 | −0.015 | −1.16 | 48% | 0.49 |
| lowvol_60d | −0.023 | −1.50 | 47% | 0.46 |
| Portfolio, net | Ann. ret | Vol | Sharpe | Max DD |
|---|---|---|---|---|
| amihud | 8.5% | 8.2% | 1.03 | −13.5% |
| ml_combo | 7.3% | 9.3% | 0.79 | −9.1% |
| mom_12_1 | 2.0% | 16.0% | 0.12 | −36.7% |
| equal_weight | −7.8% | 13.7% | −0.57 | −69.1% |
I calibrated Almgren–Chriss liquidation to a live Deribit BTC-PERPETUAL book, taking volatility, volume and the touch spread from the feed and scaling impact as 1/ADV. On a 10%-of-ADV order, $41.4M over five days, it takes 38% off TWAP's cost risk and pays for it: 7.7 bps of expected impact against 4.8.
I then trained a policy-gradient agent on 207,361 real five-minute bars, cut into 5,752 three-hour episodes and split chronologically, with every input estimated only from bars that closed before the episode opened. It was never shown the closed form and converged to it anyway: 0.060 bps apart out of sample, paired t = 0.10.
| Out of sample, bps | Shortfall | Risk | Objective | σ / episode |
|---|---|---|---|---|
| TWAP | 1.499 | 0.047 | 1.546 | 49.2 |
| Almgren–Chriss | 1.623 | 0.030 | 1.653 | 39.8 |
| REINFORCE agent | 1.686 | 0.026 | 1.713 | 33.8 |
Macro regimes from 120+ FRED-MD indicators drive a sector allocation. In developed markets the best regime-aware long-only book earns 11.2% a year against the equal-weight benchmark's 5.6%, after lagging every macro input by its publication delay. Position sizing is then posed as a convex program with transaction costs inside the objective: an L1 turnover penalty, a Cholesky second-order-cone risk term and a market-impact term, re-solved monthly over a receding horizon. Each cost level takes 494 solves, and every one reaches optimality.
The impact coefficient is assumed, not fitted from tick data. So the useful test was sweeping it. Drag the chart: the multi-period line barely moves across the whole 0–100 bps range while the frictionless book collapses.
I bootstrap a discount curve from FRED Treasury constant-maturity par yields and mark a ten-swap, $255M book from 2 to 30 years. Every par instrument reprices to under 1e-9. Daily P&L splits into carry, roll-down and level, slope and curvature moves through key-rate durations, with convexity booked as an explicit residual so the parts reconcile to a full revaluation.
I then extended it to counterparty exposure, block-bootstrapping 8,478 joint daily observations of the curve and Moody's credit spreads (1986–2026) into 20,000 thirty-year paths. Peak EPE $2.35M, peak PFE $13.5M. Wrong-way risk was priced by experiment: resampling rates and credit from the same days keeps their −0.38 correlation, resampling them from different days destroys it. The coupling is worth $89,553, 16.9% of CVA, and both runs share their curve paths bit-for-bit, so the difference cannot be Monte Carlo noise.
A dependency-free C++20 engine values the same contract three independent ways: closed-form Black–Scholes with analytic Greeks, Monte Carlo with exact terminal sampling, and a Crank–Nicolson finite-difference solver on a 400×400 grid — Thomas algorithm for Europeans, projected SOR on the early-exercise complementarity problem for Americans.
They agree to roughly 0.6% on an at-the-money call, pinned as regression tests. On barriers they don't, and that disagreement is the interesting part.
| Contract | Black–Scholes | Monte Carlo | Crank–Nicolson |
|---|---|---|---|
| European call | 10.4506 | 10.4794 | 10.4481 |
| European put | 5.5735 | 5.5616 | 5.5711 |
| American put | — | — | 6.0875 |
| Up-and-out call | — | 3.5162 | 3.3326 |
I built a deterministic limit order book and matching engine in C++20 that enforces strict price–time priority. Resting orders sit in intrusive doubly-linked lists per price level, so cancelling is a pointer splice rather than a search. A pre-warmed memory pool means the matching loop never allocates — which is what keeps the tail flat, not just the mean.
Prices are fixed-point unsigned integers with no floating point on the hot path, because float comparison in a price-priority book is a correctness defect rather than a slowdown.
| Operation | Cost | Why |
|---|---|---|
| Add — existing level | O(1) avg | hash + tail append |
| Add — new level | O(log L) | ordered-map insert |
| Cancel | O(1) avg | intrusive splice |
| Execute at best | O(1) / fill | FIFO head pop |
| Best bid / ask | O(1) | map front |
The calibration recovers α, ρ and ν to within 0.01 of the truth under 30bp of injected quote noise. On Deribit's 18SEP26 expiry it fits 21 strikes at 0.43 vol points RMSE, with the worst fit in the 95,000-strike wing.
There are no labels, so it ranks rather than classifies. Isolation Forest and ECOD over 27 order-flow features agree on 76% of their top-5% flags; the ensemble exists for the other quarter.
Across WTI and Henry Hub settlements from EIA (1990–2024), the detector flags 112 events, 79 of them corroborated by an Isolation Forest. It had been blind to the −$37.63 WTI print because a log transform turned it into NaN.
Pairs are chosen on the first 60% of the sample only, and all 15 Engle–Granger tests go through Benjamini–Hochberg. On real Deribit data no pair survives out of sample. An earlier, better-looking result came from look-ahead and was deleted.
A Polymarket scanner for multi-outcome baskets whose prices don't sum to 1. The 2028 nominee baskets sum to about 0.93: that is not free money but the cost of capital, roughly 3.4% a year for locking collateral up to 2028.
I implemented trees, ensembles, SVM, MLP, LSTM and PCA by hand. Logistic regression, linear SVM and a decision tree match scikit-learn's test accuracy to four decimals; the random forest does not (0.937 against 0.958), and the README says so.