Martin.

Projects

A collection of projects I've built and am planning to build.

Built

StratumLSM

Live

Write-Optimized Storage Engine

A Log-Structured Merge-Tree storage engine in C++ with a SkipList-based MemTable, CRC32-protected Write-Ahead Log for crash recovery, and size-tiered compaction. Integrates with the Linux perf subsystem to quantify write amplification and hardware-level execution metrics.

47,100 ops/sec sequential writes 230,800 ops/sec random reads ~3.0x write amplification
C++ CMake Linux perf CRC32 SkipList MVCC

VectorBench

Live

Columnar Execution Engine

Benchmarks SIMD-vectorized operators against scalar tuple-at-a-time processing. Combines AVX-512 16-wide parallelism, Apache Arrow-inspired memory layouts, and dictionary encoding to measure the throughput ceiling of columnar query execution.

15.5x speedup with dictionary encoding 12.8x speedup with AVX-512 SIMD 2.7x speedup with Arrow layout
C++ AVX-512 Apache Arrow CMake Linux perf

PageForge

Live

Persistent Storage Engine

A C++ storage manager combining a buffer pool with Clock replacement, a disk-resident B+ Tree index, and a slotted-page heap file for variable-length records. Uses a custom PageGuard RAII wrapper to ensure pin safety across exceptions and automatic dirty-page flushing on eviction.

O(log n) search and insert via B+ Tree Zero external fragmentation via slotted pages O(1) Clock eviction approximating LRU
C++ CMake RAII B+ Tree Buffer Pool

QueryScope

Live

SQL Query Optimizer & Plan Visualizer

Rule-based SQL optimizer using sqlglot to parse queries into ASTs and apply composable rewrite passes: predicate pushdown, projection pruning, and constant folding in a fixed-point iteration loop. Paired with a React frontend that renders before/after logical plan trees as annotated node graphs with per-pass telemetry.

29.7% AST complexity reduction 91 to 64 node reduction on complex query TPC-H semantic correctness validated
Python sqlglot FastAPI React PostgreSQL TPC-H

C++ Conductor

Live

Fixed-Size Thread Pool Library

A high-performance thread pool library built from scratch in C++ using mutexes and condition variables for thread-safe task queuing. Workers pick up enqueued callables and return std::future handles, making it straightforward to parallelize workloads without managing threads directly. Validated with Google Test.

std::future-based async results Zero-overhead task submission Concurrent correctness via GTest
C++11 CMake GTest std::mutex std::future

LocalDev CLI

Live

Docker Environment Manager

A Go CLI tool that parses YAML service definitions and drives the Docker Engine API to spin up or tear down local development environments in a single command. Handles image pulls with progress feedback, port and environment variable mapping, and a --force flag to automatically resolve container name conflicts.

Single-command multi-service teardown Real-time image pull progress --force flag for conflict resolution
Go Docker Engine API YAML CLI

Planned

CampusDine

Planned

AI-Powered Dining Hall Meal Planner

A web app that scrapes local campus dining hall menus daily and uses AI to construct an optimal meal plan based on your nutrition goals and dietary preferences. Integrates messaging and email to send your personalized plan each morning, with a conversational AI interface for adjusting macros, swapping meals, or asking nutrition questions.

Daily menu sync from dining hall sources AI-optimized meal plans per nutrition targets Email + message delivery of daily plan
React Python FastAPI Claude API PostgreSQL SMTP

QueryForge

Planned

Vectorized Physical Query Executor

A physical query execution engine that closes the loop across StratumLSM, PageForge, and QueryScope. Takes logical plans from QueryScope and executes them against the PageForge buffer pool using vectorized operators: hash join, sort-merge join, aggregation, and selection, each with explicit memory budgets. The goal is a full database stack built entirely from scratch.

Hash join, sort-merge join, aggregation operators Memory-bounded vectorized execution End-to-end: storage through query results
C++17 CMake PageForge QueryScope StratumLSM