class ReiSamsami:
role = "Applied AI / ML Engineer"
domain = ["AEC", "construction analytics", "inspection", "infrastructure", "BIM"]
toolkit = ["Python", "SQL", "Databricks", "PyTorch", "scikit-learn", "Streamlit", "OpenAI API"]
focus = "turning messy field, document, and model data into useful AI products"I build applied AI, ML, and analytics systems for construction, inspection, infrastructure, and field operations.
Assistant Professor at the University of New Haven
Ph.D. in Civil Engineering | M.S. in Data Science
My work sits at the intersection of AI engineering, data products, and built-environment workflows: multimodal inspection tools, document intelligence copilots, SQL dashboards, BIM/lakehouse analytics, risk prediction, and agentic AI systems.
| Project | What it demonstrates | Stack |
|---|---|---|
| BIMOps AI: Revit to Databricks Lakehouse | Revit schedule exports transformed into Bronze, Silver, and Gold lakehouse layers with BIM inventory, program analytics, MEP/structural summaries, and metadata readiness scoring. | Databricks, Delta Lake, SQL, Python, Revit, BIM |
| Construction Safety SQL | Relational data modeling and SQL dashboarding for hazards, incidents, near misses, corrective actions, contractor comparisons, and safety KPIs. | SQL, SQLite, Python, Streamlit, pandas |
| Project | What it demonstrates | Stack |
|---|---|---|
| Agentic Construction Safety Copilot | Multimodal safety assistant that analyzes project context and site imagery to generate observations, risk summaries, corrective actions, and report-ready outputs. | Python, Streamlit, OpenAI API, multimodal AI, structured outputs |
| Construction Docs Copilot | Document intelligence copilot for technical manuals, specifications, and safety documents with grounded answers and source-backed summaries. | Python, Streamlit, OpenAI API, PyPDF, python-docx, Pydantic |
| Project | What it demonstrates | Stack |
|---|---|---|
| Inspection Report Generator | Converts field or drone photos into structured defect findings, severity assessments, GPS-aware context, and report-ready summaries. | Python, Streamlit, OpenAI Responses API, Pillow, Pydantic, geopy |
| BridgeTwin Inspector | Bridge inspection and digital-twin-oriented workflow for infrastructure condition review and visual analytics. | Python, AI/ML, infrastructure analytics |
| BridgeWatch | Bridge inspection dashboard concept for organizing condition, risk, and inspection intelligence. | Python, dashboards, inspection analytics |
| Project | What it demonstrates | Stack |
|---|---|---|
| Project Risk Predictor | ML dashboard for predicting schedule delay, budget overrun, and expected delay days with what-if analysis and explainability views. | Python, scikit-learn, pandas, NumPy, Plotly, Streamlit |
messy domain data
-> structured extraction
-> cleaned analytical layer
-> interpretable model or AI workflow
-> usable product for decision support
I like projects where the technical work has to survive contact with real users: inspectors, construction managers, engineers, researchers, safety teams, and operations stakeholders.
That usually means:
- clean data pipelines before flashy models
- structured outputs before vague AI responses
- explainability before black-box predictions
- dashboards that answer operational questions
- AI systems grounded in domain context
Selected research/project areas:
- Vision transformers and deep learning for bridge inspection
- YOLO-based thermal bridge detection from UAS imagery
- RAG and LLM workflows for construction inspection question answering
- Prompt engineering for generative AI in the AEC industry
- BIM/BrIM inspection workflows with UAS, photogrammetry, and geospatial analytics
- Operational integration of UAS data for transportation infrastructure
Full publication list: Google Scholar
I have worked on more AI, engineering, and research projects than what is visible here. Some work cannot be shared publicly because of funding restrictions, ownership, collaboration agreements, or client constraints.
GitHub is where I keep representative public projects that show the systems I build, the problems I care about, and the technical directions I am actively developing.
I make one of the best baklavas in the world, and yes, precision matters everywhere.
