Projects

Selected systems and numerical-computing work, described by the problem, my contribution, and the result.

Distributed Query Profiling

Problem

A single-node profiler could not explain latency, skew, and hotspots across a distributed execution hierarchy.

Work

Connected frontend, planning, optimization, worker, and operator metrics through stable identifiers while preserving node- and thread-level detail.

Result

Produced query/operator/node/thread diagnostics and used them to identify lock contention introduced by high-frequency instrumentation.

  • C++
  • SQL engine internals
  • MPP execution
  • Performance profiling

Fourth-Order Cut-Cell Multigrid Solver

Problem

Standard high-order stencils lose accuracy near irregular boundaries, cut cells, and multi-component interfaces.

Work

Built fourth-order finite-volume operators using cut cells, PLG polynomial reconstruction, weighted least squares, geometric multigrid, and block relaxation for irregular unknowns.

Result

Achieved approximately fourth-order convergence and near-optimal refinement complexity; a sparse bottom solve reduced one benchmark from 118.770 s to 4.441 s (26.7×).

  • C++
  • OpenMP
  • Finite volume
  • Geometric multigrid
  • Sparse linear algebra

Density-Stratified Flow Solver

Problem

High-order spatial discretization, pressure projection, time integration, and moving geometry must remain accurate and stable when coupled.

Work

Implemented a modular two-dimensional C++ solver with PLG-based finite-volume operators, pressure projection, geometric multigrid, and OpenMP-accelerated numerical kernels.

Result

Verified fourth-order accuracy in manufactured-solution studies and obtained results consistent with reference experiments.

  • C++
  • OpenMP
  • LAPACK / MKL
  • HDF5 / Silo
  • Navier–Stokes

Market-Data and Analysis Tooling

Problem

Research and live operations needed a consistent path from brokerage feeds to storage, monitoring, and ad-hoc data access.

Work

Combined brokerage API integration, concurrent C++/Python ingestion, SQLite/TDengine storage, and lightweight internal query interfaces.

Result

Delivered a unified workflow for market-data subscription, monitoring, and operational investigation without claiming unverified latency figures.

  • C++ / Python
  • SQLite
  • TDengine
  • Brokerage APIs

Company source code and unpublished research repositories are not linked publicly.