Toan "Thomas" Dao
Senior Data Scientist building production ML systems and MLOps platforms. Previously at Unilever, MoMo, and One Mount.
About
I'm a Senior Data Scientist focused on taking machine learning from notebook to production. Most of my work sits at the intersection of modelling, MLOps, and data engineering — turning research prototypes into systems that run reliably at scale.
I've led work at Unilever, MoMo, and One Mount across demand forecasting, real-time ad decisioning, and platform MLOps. I care about clear evaluation, simple architectures, and shipping things people actually use.
Outside of building, I enjoy mentoring junior engineers and writing about ML systems and MLOps practice. The blog on this site is where I work out ideas in public.
Selected Work
Production ML, MLOps, and data engineering projects across supply chain, advertising, and platform work.
EfficientDet.Pytorch
Open-source PyTorch implementation with 1.4k GitHub stars and academic collaboration.
- 1.4k stars on GitHub
- 15% parameter compression without accuracy loss
- Academic collaboration with Shanghai Jiaotong University
AI-Powered Demand Forecasting
Enterprise-grade forecasting system improving accuracy by 15% for global supply chain.
- 15% improvement in forecast accuracy
- Reduced inventory costs by $2M annually
- Processing 10M+ SKUs daily
Integrated AI Agent Platform
Supply chain platform raising service levels to 90% with automated S&OP processes.
- Raised service levels to 90%
- Reduced manual planning by 30%
- Real-time inventory optimization
Enterprise MLOps Infrastructure
Scalable ML pipeline supporting 10+ models with automated lifecycle management.
- Automated lifecycle for 10+ models
- Enabled collaboration for 5+ data scientists
- Zero-downtime deployments
Real-time Ad Decision Engine
High-performance engine improving eCPM by 40% and CTR by 60% at MoMo.
- 40% improvement in eCPM
- 60% increase in CTR
- 1200 RPS with 100ms latency
Real-time Ad Capping System
Distributed system processing 30M+ events/day with Redis and Pub/Sub architecture.
- Processing 30M+ events/day
- Prevented overspending incidents
- Enhanced ad delivery efficiency
Prophet-based Inventory Forecasting
Marketing collaboration system driving 14% increase in campaign ROI.
- 14% increase in campaign ROI
- Optimized inventory management
- Cross-team collaboration success
Get in touch
Open to senior data science and ML platform roles, freelance work, and collaborations. I usually reply within a day.
Location
Open to remote work globally.
Ho Chi Minh City, Vietnam