Hassan R.

Machine Learning Engineer | Geospatial, Forecasting & Data Pipelines

I build geospatial and forecasting systems that survive contact with real data — not notebooks that only work on the sample file. Six years of this. My day work is production geospatial engineering: territory generation, routing, spatial clustering and demand forecasting for FMCG field operations, including projects for BAT, Mars, P&G, Nielsen and Unilever across markets in South Asia, the UK, France and Saudi Arabia. On Upwork I've delivered 653 hours of ML work, including a 195-hour contract rebuilding a client's time-series classification pipeline. Selected work: • RTM Analytics Platform — architected and deployed a multi-tenant geospatial analytics platform (FastAPI, Docker, AWS S3) serving FMCG field operations across five country markets. 20+ spatial and data-processing tools that replaced manual analyst workflows. • Large-scale spatial processing — building-footprint datasets approaching one million polygons processed with DuckDB Spatial and GeoParquet: spatial joins against administrative boundaries, out-of-core rather than loading into memory. LandScan population rasters at hundreds of millions of grid cells for population-weighted clustering. • Territory and route generation — 131 territories on a UK field-force scheduler, 664 outlets across 9 territories in Paris, 780 outlets across 4 routes in Jeddah. Voronoi tessellation, convex-hull polygons, HDBSCAN/K-Means clustering, KML outputs field teams actually use. • Sales and demand forecasting — product, brand, company and territory level for FMCG and retail clients, with satellite imagery and computer vision used to identify market areas and trigger demand alerts. • Time series classification — a 195-hour Upwork contract rebuilding and validating a client's classifier on sequence data. • Speech and NLP — a deployed Wav2Vec2 mispronunciation-detection API, and an MS thesis on Urdu speech-to-text using BiLSTM and attention models. What I'm hired for most: • Geospatial data engineering — spatial joins and intersections at scale, territory and catchment generation, clustering, coverage-gap analysis, KML/GeoParquet pipelines • Time series forecasting and demand planning — sales forecasts, seasonality and changepoint modelling (Prophet, ARIMA, gradient boosting, deep learning) • Time series classification — sensor, signal and sequence data • ML data pipelines — Airflow ETL, feature engineering, and turning messy spreadsheets, scraped web data and databases into training-ready datasets I also build LLM systems: RAG chatbots over PDFs, CSVs, Excel and SQL databases; AI agents with tool use and multi-step workflows; LLM apps on OpenAI, Claude and Gemini using LangChain / LlamaIndex and vector databases; NLP classification, extraction, summarization and semantic search. Stack: Python, GeoPandas, Shapely, DuckDB Spatial, GeoParquet, GDAL, PyTorch, TensorFlow, scikit-learn, Prophet, pandas, NumPy, FastAPI, Docker, Apache Airflow, AWS S3, PostgreSQL, MySQL, MongoDB, LangChain. Track record here: Top Rated, 100% Job Success, 653 hours, and repeat clients — one came back four separate times. Tell me what you're trying to predict, map or automate. If ML isn't the right tool for it, I'll say so before you spend money.