Bishal Dhungana
26.7153° N, 80.0534° W — WEST PALM BEACH, FL

Turning imagery and LiDAR into cloud systems that run real infrastructure.

Geospatial Data Scientist working end to end: computer vision on drone imagery and LiDAR (including a switch-state classification model I independently trained and deployed to production on AWS SageMaker), cloud pipelines and infrastructure-as-code, and the AI agents that tie it together, from field survey to production platforms protecting critical utility infrastructure.

Computer VisionLiDAR / Point CloudsCloud & AWSGeospatial Data EngineeringAgentic AI
Resume ↓ Get in touch GitHub ↗ LinkedIn ↗
Simulated LiDAR scan
drag to rotate
AI Agents · LangGraph
Cloud · AWS Lambda / S3
Computer Vision · PyTorch
01 — About

Field-to-production, end to end

I work at the intersection of geospatial engineering, applied AI, and cloud data platform engineering. Currently I'm a Data Scientist and supporting product owner at Florida Power & Light, technical lead across their AI-based infrastructure inspection program: building the data pipelines and GIS applications behind it, and independently training and deploying a computer vision model to production. My background started in the field as a Geomatics Engineer in Nepal, then sharpened through flood-hazard research at Florida Atlantic University, which led to a peer-reviewed publication and ongoing work as a core contributor to PyWMP, an open-source watershed modeling framework. Outside of work I build open-source geospatial AI tools, see Featured Builds below.

beesal13dh@gmail.com FAA Part 107 — Remote Pilot
02 — Experience

Professional experience

APR 2025 — PRESENT
Florida Power & Light - Contract, West Palm Beach, FL
Data Scientist (Geospatial & AI)
Program & Vendor
  • Defined ML scope and technical specs as data scientist/supporting product owner for a multi-million-dollar AI vendor engagement; authored RFP technical requirements and served as technical SME through development and benchmarking.
  • Evaluated ~20 vendor bids with a GPT-based scoring agent, then audited 3 external ML vendors post-award for licensing compliance and supported resolving CUDA/GPU and batch-processing mismatches against AWS SageMaker.
Machine Learning
  • Independently built and deployed an overhead switch-state (open/closed) classification model, training it locally then migrating to AWS (ECR, SageMaker) for production inference.
  • Contributed to testing and evaluating CV architectures (YOLO v8/v9/11, Faster/Mask R-CNN, EfficientDet, SSD, DETR/Deformable DETR, Detectron2, MMDetection, RT-DETR) against a precision/recall/F1 and image-scale evaluation methodology co-designed with the team, producing statistical evaluation reports for leadership; selected RT-DETR for transfer learning and standardized the approach for vendors.
  • Used Roboflow for annotation and data preparation on both production CV models, including a 6x augmentation pipeline (flips, brightness, and other transforms), train/test/validation splits, and YOLO-format export.
  • Co-developed an equipment detection and segmentation model (with a data scientist and software engineer) spanning 27-30 asset classes, evolving from object detection to segmentation.
Data Engineering for AI
  • Built an API-based metadata pipeline with spatial (10m/30ft) and temporal (30-day) joins matching 350K+ condition reports to an 8M-image catalog, enriching ~2M images with condition metadata.
  • Led the pipeline's move from ad hoc Python/GeoJSON scripts to an FME-to-PostGIS pipeline structuring 8M+ images and 65M+ associated annotations, with automatic time/version tracking via FME triggers. Designed the schema and cut 100+ metadata columns to ~50 with spatial indexing.
  • Designed the annotation-storage schema for multiple model versions, and set up the taxonomy as a feature service with editor logging and concurrent multi-user updates, plus a separate read-only REST API for internal teams and vendors, replacing a fragile Excel-based process.
Product & Analytics
  • Built a Streamlit app (later migrated to Power BI) with REST APIs summarizing dataset availability and taxonomy patterns for leadership.
  • Acted as product owner for a dataset-creation plugin built with the enterprise IT team, defining filters, UI, and delivery logistics, and used it to deliver 300+ curated imagery datasets to vendors.
Platform & Infrastructure
  • Designed an AWS CDK infrastructure-as-code framework with enterprise naming, tagging, and deployment patterns used across ML and application workloads.
  • Built a data integration platform (Python, SQL, DuckDB) with a FastAPI/React console, and manage the PostGIS databases and AWS (Lambda, S3, SageMaker) architecture behind it.
Applications
  • Built and deployed GIS applications (ArcGIS Enterprise, Experience Builder, Dashboards, Utility Network) for infrastructure monitoring and disaster response, including spatial web portals used in real time during Hurricane Milton for structural impacts, outages, and crew deployment.
  • Built a storm-pilot deployment application in ArcGIS Experience Builder to manage internal and external UAV pilots, with real-time location tracking, multi-deployment coordination, time-of-deployment logging, and multi-level access for enterprise users.
SEP 2024 — MAR 2025
Florida Power & Light - Contract, Jupiter, FL
Associate Geospatial Analyst
  • Led QA/QC of LiDAR-derived power grid data for infrastructure resilience work, including pole hardening.
  • Managed PostgreSQL/PostGIS databases, optimized Cloud-Optimized GeoTIFFs, and automated QA workflows in FME, ArcPy, and Model Builder to streamline data storage and validation.
  • Built a FastAPI sidecar for heavier geoprocessing and format conversion.
  • Built a suite of LiDAR-based transmission corridor tools: clearance and right-of-way mapping (GeoPandas, pykml, ArcPy); a point-cloud classification tool extracting pole height, length, top/bottom diameter, tilt, and communication attachment points; a CLI for vegetation-density classification (low/medium/high) supporting vegetation management at FPL's Turkey Point nuclear site; and a terrain-model generation CLI (GDAL, PDAL, GeoPandas) built against corridor LiDAR.
JAN 2023 — AUG 2024
Florida Atlantic University, Boca Raton, FL
Research Assistant
  • Conducted flood risk assessment for 3 Florida communities using FEMA hydrologic and hydraulic (H&H) modeling methods and Python automation, as part of FAU's Center for Water Resiliency and Risk Reduction (CWR3) Watershed Master Planning Initiative, a $1.7M FEMA-funded effort to build a statewide Watershed Master Plan template for Florida.
  • Built an automated ArcGIS Pro geoprocessing pipeline, cutting processing time by 50%.
  • Built a Python geocoding tool processing 3 million addresses, replacing a paid service; cut turnaround from 3 months to 1 week and saved $20K in licensing.
DEC 2020 — DEC 2022
Freelance Consultant, Kathmandu, Nepal
Geomatics Engineer
  • Led survey teams for municipal road, control, and construction surveys, including GNSS landmark and UAV topographic/transmission-line surveys.
  • Designed municipal geodatabases (land use, soil, cadastral, resource maps) with topology and attribute rules.
  • Served as GIS expert for Nepal's Survey Department land-use hand-over program, training 40+ officials in spatial analysis and database management.
  • Digitized historical land records as part of a national land-use modernization effort, replacing manual paper-based archives.
03 — Featured Builds

Self-directed AI & geospatial projects

Open-source contributions built outside of my employment, focused on hazard-resilient geospatial AI and critical infrastructure. Most ship with a live demo; nothing here requires an API key to try.

Also shipped

Outside the core hazard/geospatial focus — still real, shipped work

Currently building

Private repos, in active development — details intentionally light

Live Straight from GitHub — auto-refreshes daily

04 — Systems & Platform Engineering

What sits behind the models

The Experience section above covers the program-level work; this is a few platform-layer components that don't fit neatly into a resume bullet.

Cloud Infrastructure

Cloud Infrastructure Components

AWS CDK component framework with enterprise naming, tagging, and deployment abstractions for ML and app workloads.

Computer Vision

Utility Asset Detection Inference Service

ONNX-based object detection package for utility equipment with COCO JSON output and optional visualization.

Data Engineering

Schema Migration Utility

PostgreSQL schema migration CLI with source/target connection separation, dry runs, and preflight checks.

Integration Toolkit

API + Taxonomy Integration Toolkit

Reusable package for API extraction, taxonomy joins, metadata patching, and downstream publishing automation.

05 — Toolkit

What I build with

Machine Learning & Computer Vision

PyTorchTensorFlow/KerasUltralyticsRoboflow YOLORT-DETRDetectron2 Faster/Mask R-CNNEfficientDetDETRMMDetection ResNetXGBoostOpenCVONNX

Generative AI & Agentic Systems

LangChainLangGraphRAG Model Context ProtocolFastMCPVector Databases OllamaChromaHugging Face Transformers

Data Engineering

PythonSQLETL/ELT Pipelines DuckDBDatabricksData LakesParquet

Geospatial

ArcGIS Pro / Enterprise / OnlineUtility NetworkPostGIS DuckDB SpatialGDALPDALGeoPandas ShapelyPyProjLiDAR / Point CloudOverture Maps Cloud-Optimized GeoTIFFFoliumFME

Cloud & DevOps

AWS LambdaS3SageMaker ECR / ECSAWS CDKDocker IAMVPC / Network SecurityCI/CD

Software Engineering

JavaScript / TypeScriptRFastAPIReactStreamlit

Field

GNSS / GPS SurveyUAV Survey (FAA Part 107) Differential GNSSGeodatabase Design
06 — Research & Selected Work

Publications & academic projects

Most of my professional work is confidential to the organizations I've supported; publications below are what I can share in full.

Dhungana, B.; Liu, W. Urban–Rural Exposure to Flood Hazard and Social Vulnerability in the Conterminous United States. ISPRS Int. J. Geo-Inf. 2024, 13, 339. doi.org/10.3390/ijgi13090339 ↗
One of four core contributors and one of seven named authors, PyWMP ↗, an open-source Python framework for multi-fidelity 1D, 2D, and hybrid event-based flood modeling (Florida Atlantic University).Manuscript in review
07 — Credentials

Education & certifications

MSc, Geosciences — GIS Concentration · Florida Atlantic University · 2023–2024

BSc, Geomatics Engineering · Tribhuvan University, Nepal · 2016–2021

NASA ARSET Certifications (12) — applied remote sensing training across climate risk assessment & projection, disaster & building-infrastructure risk modeling, wildfire and watershed impacts, ecosystem services, air quality (geostationary satellites & environmental justice), water quality (inland lakes & SeaDAS), soil moisture/drought, flood monitoring & modeling, and urban heat vulnerability.

Esri Training Certifications (8) — GIS Basics, Cartography, Spatial Data Science, Imagery Analysis, Spatial Analysis, GIS for Climate Action, GIS/BIM integration for AEC, and deep-learning object classification in ArcGIS Pro.

08 — Contact

Let's talk about what you're building.

Open to Data Science, GIS/Geospatial Engineering, and Data/Platform roles. Based in West Palm Beach, FL.

beesal13dh@gmail.com linkedin.com/in/dbishal github.com/DBishal13