Muneeb ul Hassan.

Designing, building, and deploying production machine learning systems for French real estate.

• Built the production Automated Valuation Model (AVM) — an attention-based comparable-sales neural network (PyTorch Lightning) that estimates property price as a learned, attention-weighted aggregate of comparable transactions and listings, with bounded per-comparable price adjustment for built-in interpretability.
• Designed the uncertainty-estimation layer using Conformalized Quantile Regression for calibrated prediction intervals and confidence scores derived from the model’s internal signals.
• Developed the DPE/GES energy-rating estimation service — gradient-boosted models (XGBoost) with surface-aware conversion to the regulatory A–G class and a custom asymmetric loss.
• Engineered spatial neighbour-based feature pipelines using similarity-weighted aggregations of nearby properties’ known ratings.
• Contributed to Scan’Immo computer-vision research (YOLOv5, OneFormer, DPT, LoFTR, CLIP) toward floor-plan reconstruction from photos.
• Built and operated the ML infrastructure — BentoML services, MLflow tracking/registry, and notebook-driven experimentation with A/B evaluation.

• Worked on Explainable Deep Learning for Multimedia Indexing and Retrieval and conducted a survey on Explainable AI.
• Proposed a technique to justify classification decisions; paper accepted at the CBMI 2019 conference.