Electrical & Computer Engineering student at The University of Texas at Austin and a Software Engineer Intern at Amazon. I build across the stack with a focus on low-latency C++ trading systems, quantitative research (options pricing & market microstructure), distributed systems, applied ML + full-stack web, and formally-verified hardware (SystemVerilog / FPGA). I like shipping things that actually run: most of my projects have a live demo, reproducible benchmarks, or a green CI badge — not just a README.
🎯 Seeking Summer 2027 Software Engineering / Quant-Dev internships — low-latency systems & applied ML. 📍 Austin, TX
Each ships with a real test suite and green CI; several have a live demo or fully reproducible benchmarks you can run yourself.
| Project | What it is | Built with |
|---|---|---|
| nanobook | C++20 NASDAQ TotalView-ITCH 5.0 limit order book + price-time matching engine, engineered like HFT infra: a zero-allocation hot path (proven by a counting-allocator test), O(1) top-of-book via an occupancy bitmap, and a lock-free symbol-sharded pipeline — verified against a std::map oracle over millions of messages, 67.7M-execution fuzzing, ASan/UBSan/TSan, ~99% coverage, and a real 268M-message trading day. |
C++20 · CMake · GoogleTest · libFuzzer |
| vol-lab | Options-pricing & implied-volatility-surface engine on real Deribit data — four cross-verified pricing engines (Black–Scholes, CRR binomial, Monte Carlo, Longstaff–Schwartz American), SVI smile calibration under no-arbitrage (Durrleman g(k) ≥ 0 butterfly gate), and an IV solver that returns None on vega-collapse instead of a plausible-but-wrong vol. 201 property-based tests, 96% coverage, mypy-clean. |
Python · NumPy/SciPy · Hypothesis |
| micro-lab | Market-microstructure research platform: independent NASDAQ order-book reconstruction validated to 100% exact transitions over 1,673,089 events, a from-scratch statistics engine golden-tested to 1e-8, purged/embargoed walk-forward CV, and Benjamini–Hochberg FDR control over a pre-registered 17-test family — honest statistics, no PnL claims. 160+ tests, ~97% coverage (≥90% gated). | Python · pandas · statsmodels · scikit-learn |
| quorum | Raft consensus + replicated key-value store in Go — a deterministic-simulation core stress-tested over 140,000 adversarial fault schedules (message loss, partitions, crashes), checked for linearizability with Porcupine plus Raft invariant monitors, and go test -race clean — Jepsen / FoundationDB-style verification rigor. |
Go · Raft · Porcupine · DST |
| SystemVerilog FIFO Verification Suite | Sync, async (dual-clock CDC) & AXI4-Stream FIFOs formally proven with SymbiYosys (BMC + k-induction + liveness), Verilator-simulated to 100% coverage closure, and FPGA-characterized on ECP5 + iCE40. | SystemVerilog · SVA · SymbiYosys · Verilator · Yosys/nextpnr |
| ASL Sign-Language CNN · 🌐 live | Real-time American Sign Language recognition running 100% in-browser (MobileNetV2 → ONNX Runtime Web + MediaPipe) behind a full PyTorch training/eval pipeline, with a cross-language TS↔PyTorch numeric parity gate (~5e-7) enforced in CI. Ships an honest cross-dataset accuracy story — 55.5% (59.8% on static A–Y), not the leakage-inflated 96.9% same-dataset figure. | PyTorch · ONNX · Next.js · TypeScript |
More projects — systems, full-stack, finance & iOS
| Project | What it is | Built with |
|---|---|---|
| governor | Production-style Zephyr RTOS telemetry-and-control node, verified like safety-critical software: HAL-abstracted drivers, a 100 Hz PID loop, a CRC-framed reliable UART link (ARQ + dedup), and an explicit INIT→RUN→DEGRADED→SAFE_STOP safety state machine — proven with a 16-scenario fault-injection matrix, 178.8M-run libFuzzer+ASan on the parser (zero findings), property tests, 98% coverage, and CI on simulated hardware (QEMU + Renode STM32). |
Zephyr · C · Renode · QEMU · libFuzzer |
| Aquify ATX · 🌐 live | Full-stack map of 30+ Austin water fountains with Firebase auth, community reviews, and a climate-driven hydration recommendation engine using live weather. 340+ tests, CI green. | React · Vite · Firebase · Leaflet |
| Travel AI · 🌐 live | LLM travel-itinerary generator: async REST API with a preference→prompt recommendation engine, response caching, rate limiting, and a multi-step React UI. | FastAPI · React · SQLAlchemy · Docker |
| proforma | M&A merger model on a real announced deal (Synopsys/ANSYS) reconstructed from SEC filings — a live-formula Excel model differentially verified to the cent against a from-scratch Python engine, XBRL tie-out to disclosed facts, and an M&A memo. 300 tests, 97.3% coverage. | Python · openpyxl · XBRL |
| Edutok | TikTok-style AI learning feed: Gemini-generated flashcards with spaced repetition, gamification, and a never-empty resilience guarantee. 74 tests, CI. | SwiftUI · Gemini · Firebase |
| SafeWalk | An AI safety companion that checks in on you while you walk alone at night and escalates to emergency contacts if you stop responding. | SwiftUI · Gemini · Core Location · MapKit |
| PomodoroTimer | A polished Pomodoro focus-timer with customizable sessions, a focus-mode lock, stats, and a Firebase leaderboard. | SwiftUI · Firebase |
