GL Greg Lapinski

Principal Software Engineer · systems foundations & agentic infrastructure

Systems foundations. Agentic era.

I build runtimes, database engines, distributed backends, and developer infrastructure—then apply that systems depth to making AI agents capable, observable, and safe.

Explore the foundation CV European markets CV
Foundation
Systems · databases
Production
Distributed scale
Frontier
Agentic AI
Greg Lapinski on a snowy mountain summit
OUTSIDE THE SYSTEM Perspective helps. WINTER ABOVE THE CLOUD LAYER

01 Engineering foundation

Three disciplines.
One systems view.

Systems programming, distributed systems, and databases are not separate lanes. The hardest work happens at their boundaries—and those same boundaries now determine whether AI agents can do durable work.

03 / 05

Databases

SQL Server internals, SqlCLR hosting, type systems, execution plans, indexing, and production data performance.

Move to inspect the foundation

02 Selected systems

Work below
the interface.

The most consequential product work is often invisible. These are systems where architecture, debugging and delivery had to move together.

01 Salesforce / AsyncAPI

2021—Now

Shared infrastructure, made predictable

Turn queue contention into fair flow.

Reworked pipeline and queue internals, then introduced probabilistic throttling so workloads share capacity fairly without manual intervention.

  • C#
  • .NET 8
  • SQL Server
  • WinDbg

02 Microsoft / SQL Server

2013—2021

Eight years inside the engine

A database engine is a systems platform.

I worked across runtime hosting, Linux isolation, execution performance, diagnostics and product tooling—often owning the path from architecture to adoption.

01 / Linux engine port

Technical lead, Linux Extensibility

Ported SQL Satellite and the SQL Server type system to Linux; built namespace- and cgroup-based process isolation.

02 / Runtime hosting

SQLCLR ownership

Owned the .NET runtime integration layer hosted inside the SQL Server engine.

03 / Performance

19× · 15× · 10× · 2×

Spatial, TVP, XEvent reader and 300M-row spatial-index improvements.

04 / Debugging platform

Expert diagnosis, productized

Generated C# tooling from C++ PDBs; adopted as the standard across SQL Server.

05 / Learned optimization

MLOS / DEEM

Collaborated on learned engine tuning and co-authored the DEEM 2020 paper.

06 / Product + tooling

From remote debug to SSMS

Remote .NET debugging on Linux, BDC data pools, compare plans and plan layout.

  • C++
  • C#
  • Linux
  • .NET
  • gRPC
  • WinDbg
  • PowerShell

03 Trajectory

Twenty years
below the surface.

From speech synthesis on early mobile hardware to cloud-scale backend systems. Different surfaces; the same curiosity about what happens underneath. The newest stretch is different: with agents in the loop, it accelerates.

Twenty years of compounding. Agents just changed the slope—I work at the new pace, daily.

2021—Now

Salesforce

Principal Backend Software Engineer

Async pipelines · resource allocation · .NET modernization

2013—2021

Microsoft / SQL Server

Principal Software Engineer

Engine internals · Linux · runtime hosting · debugging

2009—2013

Microsoft / Azure

Software Engineer in Test

DNS · fault injection · deployment · load balancing

2008—2009

Microsoft / Windows Phone

Software Engineer

Network traffic · cost engine · NDIS

Earlier systems, 2001—2008Expand the origin storyCollapse the origin story +
2007—2008

EMCC Software

BlackBerry push email for Sony Ericsson on Symbian / UIQ.

2005—2007

Ideal Host Poland

Led a line-of-business platform for the UK plumbing industry.

2004—2005

Axiom Software

Co-founded and shipped an ERP-integrated sales platform.

2001—2003

IVO Software

Speech synthesis and accessibility on Symbian S60.

04 Selected open source

Agents need
systems, too.

Built from the systems side of AI: navigation, orchestration, persistence, review, and safety infrastructure that lets coding agents do durable work. Not a side quest—these tools sit in my own loop, every working day.

01 / FEATUREDRust · React · TypeScript ↗

Operating environment for AI coding agents

TermAl

Supervise persistent coding sessions from one workspace. Delegate bounded work to child agents, coordinate primary sessions through durable mailboxes and shared project state, then fan results back in.
01Persistent sessionsKeep several coding agents live across workspaces and restarts.
02Delegation + fan-inSpawn focused child work and resume when one or many results return.
03Durable mailboxOrdered agent-to-agent messages with receipts and acknowledgements.
04Shared boardsVersioned project state for agents; pinned decisions and outputs for you.
Codex · Claude Code · Gemini CLI · Cursor Agent · OpenCode
02C# · F# · MCP ↗

Local code navigation for AI agents

PhoenixCodeNav

A fast, structured map of very large C# and mixed C#/F# repositories. Phoenix keeps one local index warm, narrows each question to the projects that matter, and returns compact evidence labeled by confidence and freshness.
01Ranked repository indexPersistent SQLite search and syntax maps across C#, F#, Markdown and SQL.
02Exact C# semanticsLazy Roslyn clusters resolve definitions, references, implementations and callers—without loading the whole solution or running MSBuild.
03Ownership + dependency graphMap compile ownership, dependency direction, impact and related tests across legacy and SDK-style projects.
04Bounded agent contextReturn outlines, precise source spans and context packs instead of whole-file dumps.
Codex · Claude Code · any MCP client  /  local shared daemon
03Rust ↗

Thread-per-core async I/O runtime

Starfish

One thread-confined reactor per core, with cooperative scheduling, core-local state and lock-free cross-reactor handoff.
EXECUTIONOne reactor per OS threadLocal active queues, timers and I/O completions.
MEMORYNUMA-aware allocationCore-local ownership with MiMalloc-backed heaps.
HANDOFFLock-free cross-reactor queuesRemote messages enter the owning reactor without a shared run queue.

05 First principles

Before distributed systems,
there were brain waves.

1998—2003

Gdansk University of Technology

M.S. in Computer Science & Engineering

B.S. in Computer Science

  • Minor in Biomedical Engineering
  • Thesis: EEG analysis using wavelet transform

06 Start a conversation

Bring me
the hard part.

Systems runtime, database engine, distributed backend, or agentic developer platform—if the problem crosses layers, we should talk.

CV ↓ European markets CV ↓ GitHub ↗ LinkedIn ↗