Case studies / Forecasting
Forecasting
Predicting demand, prices, arrival times and fraud.
Approach
8 write-ups
Tag:
-
AmazonNewForecasting
Mitra-v2: Scaling synthetic pretraining for tabular foundation models
Problem Creating a new model for every table required lots of engineering and tuning.
Solution Trained a foundation model on millions of synthetic tables so it can predict missing values on any new table without extra training.
Top performance on TabArena and ScoringBench benchmarks
Read on amazon.scienceDetails -
GoogleForecasting
TimesFM-3: A zero-shot foundation model for multivariate forecasting
Problem Users needed accurate forecasts that consider many related signals, but existing models only used single series data.
Solution They built TimesFM‑3, a 330‑million‑parameter transformer pre‑trained on over 1 trillion time points, that can forecast multiple series together in one pass without fine‑tuning.
Research behind this, on AI Papers DailyChronos-2: A foundation model for multivariate time series forecastingRead on research.googleDetails -
AirbnbForecasting
How we knew COVID was over (and what our models had to unlearn)
Problem Forecasts kept missing actual demand after COVID, causing biased planning decisions.
Solution They created a process to decide whether to refit, redesign, or keep a model unchanged, analyzing bias before updating forecasts.
Read on medium.comDetails -
Hugging FaceForecasting
NVIDIA Cosmos-H-Dreams: Bringing Real-Time Generative Simulation to Surgical Robotics
Problem Testing surgical robot actions is slow, costly, and risky on real hardware.
Solution Created a real‑time generative simulator that predicts video outcomes from robot motions, using a distilled student model running on one RTX GPU.
Research behind this, on AI Papers DailySurgPace: Complexity-Aware Progressive Distillation for Surgical Multimodal Large Language ModelsRead on huggingface.coDetails -
LyftForecasting
Predicting Rider Conversion in Sparse Data Environments with Bayesian Trees
Problem Rider conversion rates were hard to predict because data for specific contexts was extremely sparse.
Solution They built a Bayesian tree model that shares information across similar contexts, avoiding over‑fitting and running fast enough for live predictions.
Read on eng.lyft.comDetails -
GoogleForecasting
Protecting cities with AI-driven flash flood forecasting
Problem Cities lacked early warnings for rapid flash floods, putting residents at risk.
Solution Created an AI model trained on news and weather data that predicts urban flash floods up to 24 hours ahead, enabling earlier alerts.
Up to 24‑hour advance notice
Research behind this, on AI Papers DailySkillful and realistic precipitation forecasting with reinforced diffusionRead on research.googleDetails -
ZalandoForecasting
Paper Announcement: A Practical Approach to Replenishment Optimization with Extended (R, s, Q) Policy and Probabilistic Models
Problem Zalando needed to avoid stock-outs and excess inventory while handling uncertain demand.
Solution They built a simulation‑driven engine that uses probabilistic demand forecasts, extended reorder policies, and Monte Carlo runs to pick robust replenishment plans.
Read on engineering.zalando.comDetails -
GoogleForecasting
Reducing EV range anxiety: How a simple AI model predicts port availability
Problem EV drivers worry about finding an open charging port, causing range anxiety.
Solution Created a lightweight linear regression model that predicts the chance a charging port will be free in the next minutes, using few real‑time features for fast, low‑latency answers.
Read on research.googleDetails
No write-ups match these filters.