← Back to case studies
Confidential Client
ML Production Platform — Forecasting, Recommendations & NLP Classification
City markets
20+
Products classified
200k+
Taxonomy maintenance
manual
→
automated
A large food delivery platform operating across 20+ cities required a production-grade ML platform to improve operational efficiency through forecasting, personalized recommendations, and automated product classification.
What changed
Before
Forecasting
Heuristics
Taxonomy
Manual
200k+ products
Recommendations
None
After
Forecast
Prophet · XGBoost
demand + ETA
Classify
TF-IDF NLP
200k+ products
Recommend
Collaborative filtering
Scope
20+ city markets
in production
Three model families — forecasting, recommendation and NLP classification — running as one production system across every market.
What we built
The client operated a high-volume food delivery platform and needed ML models in production — not just prototypes. Requirements spanned forecasting, product intelligence, and recommendations, all deployed and monitored at scale across multiple geographic markets.
Key Features
Preparation time forecasting using xgBoost and Deep Neural Networks for real-time ETA accuracy.
Order volume forecasting across 20+ cities using Facebook Prophet with COVID and weather external regressors.
NLP-based product classification at scale — 200k+ SKUs categorized automatically.
Recommendation systems using TF-IDF, collaborative filtering, and user clustering.
Full ML lifecycle: model training, evaluation, deployment, and automated retraining pipelines.
REST API delivery via FastAPI and Spring Boot for operational system integration.
CI/CD for ML models with automated testing and rollback capabilities.
Results
Production-grade ML covering forecasting, recommendations, and NLP classification across 20+ city markets.
ETA accuracy improved significantly, directly impacting customer satisfaction scores.
Automated classification of 200k+ products eliminated manual taxonomy maintenance.
Personalized recommendations increased order conversion metrics.
Stack
Engagement: 4 years · Real-time & Industrial Data Platforms, Product Engineering, AI Infrastructure & MLOps