Skip to content

Experience

Engineering practice first, then AI and digital delivery.

A year of structural and façade delivery, an industry project with AECOM, and applied R&D in inspection AI, CAD-to-BIM drawing AI and Scan-to-BIM.

  1. AI & digital

    R&D AI-Construction Specialist

    AGECS · Remote

    Apr 2026 – Present

    Applied R&D turning structural drawings and laser scans into structured, BIM-ready data.

    • Column detector on plans: mAP50 0.941, precision 0.944, recall 0.935 (YOLOv8m-seg, tiled inference, 4-rotation TTA).
    • 8-class structural-element segmentation on ~21k tiles: validation mAP50 0.902 (box) / 0.893 (mask).
    • Human-in-the-loop Scan-to-BIM: a ~264 M-point scan to a verified IFC of 2 slabs, 9 walls and 21 columns.
    • Dataset engineering across six public sources and QA'd annotation of project sheets in CVAT.
    • Evaluated multimodal LLM/VLM behaviour on engineering drawings and documented failure modes.
    PyTorchYOLOv8-segOpenCVOpen3DIfcOpenShellCVATCase study: CAD-to-BIM AI →Case study: Scan-to-BIM →
  2. AI & digital

    Co-Founder & CTO

    AECAI Ltd · Belfast, UK

    Dec 2025 – Present

    An AI-assisted structural inspection platform, from research models to a production web product.

    • Built the complete web console (Next.js): dashboards, assets, inspection forms and builder, photo review, reports.
    • Designed the spalling and exposed-rebar pipeline (patching, Gaussian stitching, TTA, post-processing) and parallel serverless GPU inference.
    • Evaluated models rigorously: corrected a misleading F1 of 0.14 to a true 0.901 for the rebar model, and ran a false-positive study across 23 test images.
    Next.jsTypeScriptSupabaseRunPodPyTorchU-NetCase study: AECAI inspection platform →
  3. Industry project

    Digital Design Engineer: Industry Collaboration

    AECOM × Ulster University · Belfast, UK

    Feb 2025 – May 2025

    Industry project scoped with AECOM: crack detection linked to a Revit asset model and Power BI.

    • Converted AECOM's 2D drawings into a Revit model and linked crack data via Dynamo and IFC.
    • VGG16 transfer learning roughly doubled crack-classifier accuracy over a custom CNN (36.67% → 70.0% on 30 test images), with width classes against EN 1992-1-1.
    • Power BI dashboard with an embedded 3D model, crack register, maintenance actions and a mobile view. Module mark 80%.
    RevitDynamoPythonKerasPower BICase study: Water tank digital twin →
  4. Engineering

    Structural & Façade Design Engineer (Graduate)

    National Consulting Engineers Inc. · Florida, USA (remote)

    Sep 2023 – Sep 2024

    Structural calculations and detailed drawings for façade and secondary-structure packages.

    • Delivered 115+ design packages: glass railings, curtain walls, canopies, stairs, walkways and connections in steel, aluminium and timber.
    • Designed to AISC, ASCE 7, ACI and the Florida Building Code; coordinated with architects, contractors and fabricators.
    AutoCADETABSSAP2000SAFEIDEA StatiCaRAM Connection
  5. Engineering

    Planning Engineer (Intern)

    DME Holding · Cairo, Egypt

    Jul 2023

    Planning department internship on a residential development.

    • Supported scheduling, cost tracking and BOQ updates in Primavera P6, and prepared progress reports.
    Primavera P6Excel
  6. Engineering

    Site Engineer (Intern)

    Redcon Construction · Cairo, Egypt

    Sep 2021

    Site internship (90+ hours) on the heritage-sensitive Sour Magra El-Oyoun project.

    • Supervised pile positioning and excavation checks, inspected reinforcement cages, ran slump tests, and took part in load testing of cast piles.
    Site QA/QC

Skills

Grouped by where I have used them.

No percentage bars. Each group names the work that evidences it.

AI & Computer Vision

  • Instance & semantic segmentation
  • Object detection (YOLOv8 / YOLO11)
  • U-Net · FPN · DeepLabV3+
  • Point-cloud segmentation (PointNet++)
  • Tiled inference & TTA
  • Error analysis & field validation

Evidence: Dissertation · AGECS · AECAI

Machine Learning & Data Processing

  • PyTorch · segmentation-models-pytorch
  • TensorFlow / Keras
  • Dataset engineering (COCO / YOLO / SVG, hash-verified)
  • Point-cloud processing (Open3D, ~264 M points)
  • Annotation QA (CVAT)
  • pandas · NumPy · scikit-learn · R
  • Metric design (IoU, mAP, F1, specificity)

Evidence: MSc COM779 (81%), COM736 (75%) · AGECS

Cloud & MLOps

  • Serverless GPU inference (RunPod)
  • Supabase (Postgres, storage, edge functions)
  • Hugging Face model hosting
  • Docker worker images
  • Vercel · GitHub Actions CI/CD
  • Git & GitHub

Evidence: AECAI production stack · this site

BIM, Digital Twins & BIM Level 3

  • Revit · Navisworks · AutoCAD
  • Federated models & clash detection
  • ISO 19650 · BEP / EIR · CDE
  • BIM Level 2 delivery; Level 3 principles (open IFC, cloud-connected data)
  • IFC · IfcOpenShell · Scan-to-BIM
  • Power BI digital-twin dashboards

Evidence: MSc BEN714 (73%), BEN715 (80%) · AGECS

Generative AI & Agents

  • Multimodal LLM / VLM evaluation on drawings
  • LLM features in products (Claude, Gemini)
  • AI coding agents: Claude Code, Codex, Kimi
  • Evaluation design & failure-mode reports
  • AI Fluency (Anthropic)

Evidence: AGECS LLM tests · AECAI console · certificate

Structural Engineering

  • RC, steel, aluminium & timber design
  • AISC · ASCE 7 · ACI · Eurocodes
  • ETABS · SAP2000 · SAFE
  • IDEA StatiCa · RAM Connection
  • Connection & façade detailing
  • Tekla shop drawings

Evidence: NCE 2023–24 · BSc graduation project (A+)

Programming & Automation

  • Python
  • TypeScript / Next.js
  • Dynamo · pyRevit
  • OpenCV · NumPy · SciPy
  • SQL
  • Jupyter · reproducible notebooks

Evidence: AECAI console · BIM automation

Languages: Arabic (native), English (fluent).