Available for senior and staff-level roles
Seattle, WASystems Engineer | LLVM/Clang, AI Accelerators & AI Infrastructure

Omar Habra

Compilers, AI infrastructure, and the hardware-aware edge.

I build systems for teams that cannot afford shallow engineering: compiler work, performance-sensitive platforms, accelerator bring-up, and product-facing AI systems that still have to ship cleanly.

Compiler and toolchain internalsAI platform reliabilityAccelerator-aware model bring-upProduct delivery under real constraints

Shipped In

Apple, SiFive, Stealth

Compiler teams, performance engineering, and product-facing ML delivery.

Core Depth

LLVM, Swift, TTNN

Public proof across compiler work and accelerator bring-up.

Operating Range

AWS, Azure, GCP

Multi-cloud systems with automation, retrieval, and reliability guardrails.

Current Search

Senior / Staff

Roles where systems judgment and shipping discipline both matter.

Search Brief

The hiring case in three minutes.

Less portfolio decoration, more signal about where I fit, how I work, and what a team can verify quickly.

01 / Best Role Fit

Senior and staff-level systems, compiler, or AI infrastructure roles.

My strongest work happens when low-level correctness, product pressure, and operational reality all touch the same system.

02 / Fastest Proof

A hiring loop can verify real depth in under ten minutes.

Open the Swift compiler PR, the TT-Metal pull requests, or DiffSwarm and you get concrete artifacts instead of portfolio theater.

03 / How I Operate

Calm diagnosis, readable systems, and follow-through when things get messy.

I like difficult work, but I do not like drama. The point is to make the system clearer, safer, and easier to move.

Compiler workAI infrastructureProduct deliveryRecent public proof

Selected Proof

Work you can inspect.

Four settings where the job changed, the constraints were real, and the work can still be inspected by a technical hiring panel.

Context

Mandate

Proof

Constraint

ABI mismatches and codegen regressions were expensive to get wrong.

Ownership

Bridged AST and ABI behavior, validated cross-architecture code generation, and built regression tooling for safer rollout.

Impact

Shipped interoperability work with stronger confidence across x86_64 and arm64 targets.

Tags

SwiftClangABI
Open the Swift PR

Operating Style

Outside work, same operating system.

I keep this section because it says something useful about how I work: train deliberately, review the telemetry, and stay composed when the conditions get ugly.

Why it matters

Not hobby theater. More evidence about judgment, repetition, and how I handle long feedback loops.

Race Telemetry

Real route data from actual races

LIVE
Loading GPS data...

01

After-action review through route traces, splits, and footage.

02

Pace patiently, then push when the data says the system can hold.

03

Preparation, stress tolerance, and clean retros are part of the same muscle.

Training lens

Telemetry

Route files, splits, and footage over vibes.

Temperament

Composed

Long races and bad debugging sessions reward the same calm.

Transfer

Deliberate prep

Training blocks and release plans rhyme more than people think.

Marathon finisher with friend holding medal

First marathon

Seattle Marathon

Four months of structured training from zero to finish line.

BJJ competitors on podium with medals

Purple belt

Brazilian Jiu-Jitsu

Repeated adaptation under pressure, not just one good day.

Athlete at triathlon finish line

Olympic distance

Lake Tye Triathlon

Swim, bike, and run with telemetry worth reviewing afterward.

Public Receipts

The work is inspectable.

Good hiring conversations get faster when the proof is one click away.

Open Product

DiffSwarm

A multi-agent review CLI where one model proposes bugs, another challenges them, and the final report keeps only evidence-backed findings.

Visit the product

Compiler Work

Swift-C++ interop

Public compiler contribution adding getter and setter interoperability to the official Swift toolchain.

Read PR #40842

Accelerator Work

TT-Metal vision models

MaskFormer, DPT-Large, and YOLOS-small bring-ups for novel hardware, with implementation detail and artifacts in the open.

See TT-Metal PRs

Contact

If the role needs depth and velocity, let's talk.

The best outreach includes the role, the hard part of the system, and why you think my background maps to it.