Senior AI Engineer | Multimodal & Vision Systems | LLM Reliability & MLOps

Results-driven Senior AI Engineer with 5 years of experience owning the end-to-end lifecycle of production AI and computer vision systems—from business requirement scoping and data strategy to cloud deployment and executive reporting.

My work bridges high-resolution visual modeling with generative AI—architecting multimodal inspection pipelines (Vision + VLM), leading large-scale data operations, and engineering resource-efficient serving platforms governed by strict reliability contracts and full-stack observability.

Open-source reliability layer enforcing AST-based query validation, read-only safety, and schema governance between LLMs and SQL engines.

• Deterministic AST Validation: Traverses Abstract Syntax Trees (sqlglot) to deterministically block destructive commands (DROP, DELETE, UPDATE, ALTER) and malformed syntax prior to execution.
• Self-Correcting Workflows: Implements schema-aware reflection routines that guide local and cloud LLMs to resolve schema and column mismatches autonomously.
• Strict Safety Contracts: Backed by Pydantic v2 schema enforcement and asynchronous state machines for predictable production execution.

Lightweight model-serving environment engineered for low-latency, cost-controlled LLM inference under hardware constraints.

• Serving Architecture: Deployed open-weight LLMs using vLLM, Docker, and single-node Kubernetes (k3s) with OpenAI-compatible API contracts.
• Compute Optimization: Benchmarked INT4 quantization, KV-cache utilization, and memory/latency trade-offs to prevent OOM bottlenecks under strict compute budgets.
• End-to-End Observability: Instrumented real-time telemetry, latency metrics, and LLM trace evaluation using OpenTelemetry (OTel), the Grafana LGTM stack, and Langfuse.

End-to-end production modeling, automated compliance auditing, and data strategy across large-scale municipal and satellite inspection initiatives.

• Multimodal AI Architecture: Designed two-stage automated compliance pipelines coupling real-time object detection models with downstream Vision-Language Models (VLMs) using structured prompting and strict Pydantic schemas.
• Spatial Optimization: Led training and deployment across satellite and street-level imagery (building footprint extraction, agricultural segmentation, urban asset detection), integrating custom post-processing and SAHI to boost small-object recall without increasing training compute costs.
• Data Strategy & Cloud MLOps: Directed annotation operations and QA governance across 138K+ high-resolution images (8 municipal categories), while orchestrating cost-optimized batch training and evaluation pipelines on GCP Vertex AI and Kubeflow.

• AWS Certified Machine Learning – Specialty — Amazon Web Services
• GitHub Copilot Certification (GH-300) — Microsoft / GitHub
• Microsoft Certified: Azure AI Fundamentals (AI-900) — Microsoft

• Technical Community & Advisory Lead @ DevOps Visions & Elmentor Program:
  • Engineering Enablement: Designed and delivered hands-on technical workshops for software engineers on integrating AI-assisted development workflows (GitHub Copilot across the SDLC) and cloud observability practices.
  • Program Delivery & Mentorship: Collaborated with industry leads and Microsoft MVPs to structure technical learning tracks via Azure DevOps, mentoring engineers on production Python standards, API design, and career progression.

• LinkedIn: linkedin.com/in/ayman-aboghonim
• GitHub: github.com/aymanaboghonim
• Email: aymanaboghonim@gmail.com