Omar Habra

• RedlineAI — local-first AI contract review: Built a privacy-first macOS AI workflow for legal-document review with source-grounded extraction, severity-scored findings, and human-reviewable evidence trails.
• DiffSwarm — multi-agent code review CLI: Built a Homebrew-installable PR review CLI that coordinates parallel Codex/Claude agents, tool-calling review passes, verifier evidence, and local BYOK execution.
• AI Tutoring Platform: Built production AI infrastructure across retrieval (Postgres/pgvector), Bedrock/Claude orchestration, Next.js, Terraform-managed AWS, structured-output validation, and curriculum-aware regression checks.
• GenAI Visual Processing: Improved image-pipeline latency and reliability through caching, batching, throughput tuning, and an evaluation harness for model regression detection.

• Improved RISC-V vector workload performance by tuning Clang/LLVM vectorization cost models and RVV codegen heuristics; added microbenchmarks to catch compiler performance regressions.
• Integrated LLVM analysis tooling into FreedomStudio IDE and built Python automation for toolchain builds and RISC-V ISA analysis.
• Automated cross-compilation, ISA validation, and performance benchmarking through CI pipelines with Ansible-managed environments.

• Worked on Swift–C++ interoperability for the Swift/Clang toolchain, including AST bridging, ABI integration, and name-mangling support for cross-language calls.
• Supported Apple Silicon compiler work by resolving ABI and backend issues across Swift, Clang, and LLVM for arm64 targets.
• Built cross-architecture CI and log-triage tooling for x86_64/arm64 compiler regressions across Clang, Swift, and LLVM workflows.

Built a Homebrew-installable AI developer tool that reviews local diffs and GitHub PRs with Codex or Claude Code, plans focused review passes, verifies findings, and outputs actionable markdown reports while keeping code local through a BYOK model.

Building a macOS legal AI workflow that reviews contracts locally with Codex or Claude Code, preserves evidence anchors, verifies findings against source text, and produces reviewable work product for NDAs, vendor agreements, DPAs, MSAs, and SOWs.