Makesense MLOps Pipeline
End-to-end multimodal MLOps pipeline for entity extraction, relationship discovery, and knowledge graph construction.
Makesense MLOps Pipeline presents a systematic study of transformer-based architectures for abstractive dialogue summarization, with emphasis on computational efficiency and deployment readiness.
Research Contributions
- Systematic Architecture Comparison — Comprehensive evaluation of encoder-decoder variants (BART, T5, Pegasus) on dialogue summarization with matched compute budgets.
- Efficiency-Quality Tradeoffs — Quantitative analysis of the Pareto frontier between ROUGE scores and inference latency.
- Training Strategy Ablations — Investigation of warmup scheduling, gradient accumulation, and mixed-precision training.
- Production-Ready Pipeline — End-to-end MLOps implementation with experiment tracking, model versioning, and reproducibility guarantees.
Links: GitHub