Building robust and scalable systems, not brittle scripts.
📍 Beirut, Lebanon | 🌍 Open to Global Remote & Relocation

I am a Software Engineer specializing in backend architecture, data engineering, and autonomous AI systems. My work focuses on translating complex business problems into robust technical solutions. I prioritize strict data validation, clear architectural boundaries, and measurable business impact. Whether I am engineering a deterministic AI compiler or building production ETL pipelines that drive executive strategy, my goal is always to write testable and maintainable code.

🧠 Languages
⚙️ Backend
🤖 AI & Data
🛠️ DevOps

The Problem: Standard ReAct agents hallucinate on numerical computations and pose severe security risks via unbounded code execution.

• System Architecture: Engineered a 2-Pass LLM Compiler that strictly decouples logical reasoning from JSON syntax formatting. The system translates natural language into a strict Pydantic V2 Intermediate Representation (IR). A deterministic Pandas and NumPy backend handles all physical execution. The LLM decides what to compute and the engine decides how.
• Zero-Knowledge Profiling: Built a deterministic pre-flight profiler that extracts regex skeletons, monotonic ID flags, and exact categorical matches. This grounds the LLM in ground-truth data without flooding the context window.
• Hermetic State & Safe Math: Implemented an immutable JSON-safe state boundary using recursive Pydantic validators to scrub C-types into native Python primitives. Replaced unsafe Python eval() with numexpr for zero-trust mathematical execution.
• Self-Healing Execution: Designed an internal correction chain powered by difflib. It traps validation errors, uses fuzzy string matching for high-cardinality data, and feeds actionable tracebacks to an isolated LLM loop to repair payloads mid-flight.
• Core Stack: Python LangGraph Pydantic V2 Pandas NumPy Qdrant Streamlit

The Problem: A manufacturing client possessed 8 months of siloed and corrupted operational data with no visibility into material substitution profitability.

• System Architecture: Engineered a multi-stage Python ETL pipeline to ingest, normalize, and validate 8 operational workbooks. Diagnosed and patched a silent data corruption bug in Excel formatting alignment that was destroying 1:1 data mapping.
• Data Modeling: Designed 7 canonical summary tables and interactive dashboards to provide self-service analytics for non-technical stakeholders.
• Metrics: Analyzed 1,880 substitution events and proved 80.9% of substitutions were cost-saving. Identified a Pareto concentration where just 3 machines drove 92% of machine-attributed cost increases. This completely reframed the client strategy.
• Core Stack: Python Pandas openpyxl Plotly Cloudflare Pages

The Problem: Fitness applications typically rely on basic CRUD patterns that fail at scale, lack data integrity, and are difficult to test.

• System Architecture: Architecting a strict Hexagonal Architecture using a Service-Repository pattern. Features a Pydantic Firewall for mass-assignment prevention and NIST-aligned security with HaveIBeenPwned k-anonymity checks.
• Data Integrity: Implemented GDPR-compliant soft-delete anonymization to preserve historical analytics while removing PII. Utilized an Extensible Catalog Pattern to seamlessly merge global system defaults with user-generated custom items.
• Status: Currently undergoing a major local refactor to the modern Python ecosystem (APIFlask, uv, Python 3.13). The modernized architecture will be pushed to my primary GitHub upon completion.
• Core Stack: Python APIFlask SQLAlchemy 2.0 Pydantic V2 Hypothesis Pytest

• 🔭 Expanding Scrygent with advanced Retrieval-Augmented Generation (RAG) and semantic memory replay using Qdrant.
• 🏗️ Refactoring Styx to a modern 2026 Python ecosystem (APIFlask, uv, Python 3.13 type parameters).
• 📚 Deepening expertise in distributed tracing (LangSmith) and serverless vector databases for autonomous agent architectures.