Thanos AI Engine.
Company
Cowlar Design Studio
Role
Lead Full Stack & ML Engineer
Timeline
2023 - 2024
Links
Drastically reduced lifecycle from raw capture to production
Highly validated via gamified Human-in-the-Loop review
Near zero-latency routing of low-confidence assets
Major Features Highlight
End-to-end ML pipeline from data to deployment
Easy-to-use annotation tools with AI assistance
Track annotator performance in real time
Gamified leaderboards to boost productivity
Collaborative platform for teams
Auto-scaling for any project size
Seamless integration of human and AI workflows
Fast setup and deployment from day one
Handles large datasets with speed and accuracy
Perfect for data scientists, ML engineers, and teams
Overview
Thanos is a custom Human-in-the-Loop AI orchestration and automated ML training lifecycle engine. It intelligently filters low-confidence inferences to human annotators while dynamically incorporating high-confidence results to retrain edge models.
The Challenge
Deploying edge models requires millions of clean labeled coordinates. Traditional labeling pipelines were slow, manual, and decoupled from model training, introducing massive friction and delays when updating models.
The Engineering Solution
Designed a consolidated orchestration suite. Engineered dynamic browser-based labeling interfaces using Vue.js. Created intelligent backend confidence routing engines in Node.js, and implemented a gamified leaderboard dashboard to optimize annotator workflows.
Key System Deliverables
- ✦Intelligent routing logic driven by Python, TensorFlow, and PyTorch
- ✦Robust backend built with Node.js, Express, and MongoDB clustered databases
- ✦Responsive Vue.js frontend featuring rich canvas-based annotation modules
- ✦Scalable Dockerized microservice architecture deployed in Kubernetes clusters
Measurable Success
- ✓Accelerated client model production timelines by 40%
- ✓Optimized annotations using collaborative gamification methods
- ✓Reduced manual overhead data cycles through automated high-confidence tagging