Portfolio · 2026

OmarHabra

CompilersAI InfrastructureEndurance

◆ Apple · SiFive · Tenstorrent ◆

Scroll

II — The Work

Compilers, AI infrastructure, and the systems underneath.

Swift–C++ interop shipped at Apple. 2× RISC-V vector speedups at SiFive. Now building production LLM platforms and shipping open-source model bring-ups for AI accelerators.

2×

RISC-V Speedup

SiFive · LLVM RVV codegen

1,000+

Apple Projects

Adopting Swift–C++ interop

500+

Concurrent Users

Production LLM tutoring

40%

Latency Cut

GenAI inference pipeline

Featured Projects

01 / 07LegalTech · Shipping

RedlineAI

$399/mo SaaS

Privacy-first macOS app for AI contract review. Flags risky clauses with severity scoring; every finding verified against source text.

Used daily by startup legal teams for NDAs and vendor contracts
Source-grounded verification on every finding
Severity-scored clause risk with jump-to-evidence UX

macOSLLM$399/mo SaaSPrivacy-first

View

02 / 07DevTools · Hacker News Featured

DiffSwarm

HNFeatured

Homebrew-installable multi-agent PR review CLI. Parallel AI agents hunt concurrency, security, and logic defects; a judge agent cross-verifies.

Runs locally via BYOK architecture — no code leaves your machine
Specialist agents propose; verifier agents refute with diff-anchored evidence
Only consensus issues with evidence make the final report

PythonMulti-AgentCLIBYOK

View

03 / 07EdTech · Contract

AI Tutoring Platform

500+Concurrent users

End-to-end ownership from concept to 500+ concurrent users: RAG pipeline, LLM orchestration, Next.js frontend, multi-tenant AWS.

99.5% answer stability across curriculum updates via Pydantic structured outputs
70% faster deploys with Terraform-managed multi-tenant AWS
pgvector RAG pipeline with Bedrock/Claude orchestration

RAGBedrock/ClaudepgvectorTerraform

04 / 07Open Source · 2025

Tenstorrent TT-Metal

3Merged bring-ups

TTNN model bring-ups for modern vision architectures on Tenstorrent Wormhole accelerators. End-to-end pipelines with validation and perf artifacts.

MaskFormer Swin-B hybrid TT decoder + perf artifacts (PR #32335)
DPT-Large (MiDaS 3.0) depth estimation bring-up (PR #33123)
YOLOS-small object detection bring-up (PR #32500)

TTNNTT-MetalComputer VisionPyTorch

View

05 / 07Retail · Contract

GenAI Visual Processing

40%Latency reduction

Latency and reliability work on a production GenAI image pipeline — intelligent caching, throughput tuning, and regression-catching evaluation.

Intelligent caching + throughput tuning on the inference path
Evaluation harness caught model regressions before they reached production

GenAIComputer VisionModel Optimization

06 / 07SiFive · 2022 – 2023

RISC-V Vector Codegen

2×Vector speedup

Tuned Clang/LLVM vectorization cost models and RVV codegen heuristics for RISC-V. Built the microbenchmark suite that prevents regressions.

Tuned vectorization cost models for RVV
Microbenchmark suite gating every compiler release
FreedomStudio IDE analysis tooling + Ansible CI

LLVMClangRISC-V RVVCodegen

07 / 07Apple · 2020 – 2022

Swift ↔ C++ Interop

1,000+Internal projects

Shipped Swift–C++ interoperability across the Swift/Clang toolchain — AST bridging, ABI integration, name mangling — adopted by 1,000+ internal projects.