Manish Tiwari


Current Projects & Roadmap

What I'm building and learning right now — focused on open source, ML systems, and deployment pipelines.

Open Source Contributions

Actively contributing to:

  • sktime — Python library for time series machine learning, unified API for forecasting, classification, and more.
  • chaoss/augur — Open source project for measuring community health and sustainability metrics.

Recent PRs: first PR on sktime, My PRs on CHAOSS/augur

Main skills: Python, unit testing, open source workflows, CI/CD.

Production Model Drift Monitoring System

Feb 2026 | GitHub | Live Demo

Designed and built an end-to-end MLOps prototype to monitor model drift in production-like settings. The system tracks changes in prediction distributions over time and triggers alerts when significant drift is detected.

  • Developed FastAPI service for model inference with logging of predictions and timestamps
  • Implemented drift detection using statistical methods (PSI, KS test)
  • Built interactive Streamlit dashboard with Plotly to visualize drift trends in real time
  • Integrated Discord webhook alerts for threshold-based drift notifications
  • Containerized application using Docker and set up basic CI using GitHub Actions

Tech Stack: FastAPI, Streamlit, Scikit-learn, Docker, GitHub Actions, SQLite, Plotly

Note: This is a prototype system focused on demonstrating end-to-end monitoring. Future improvements include better scalability, feature-level drift tracking, and performance benchmarking.

Custom Language Model Training

Jan 2026 | GitHub | Live Demo | Dataset | Model

Trained a lightweight 4.2M-parameter language model using PyTorch to understand the end-to-end training workflow, including data preparation, model architecture, and hyperparameter tuning.

  • Implemented training loop, loss tracking, and basic evaluation
  • Experimented with hyperparameters such as learning rate, batch size, and sequence length
  • Analyzed training behavior using loss curves and validation metrics

Currently exploring deployment aspects such as inference pipelines, latency, and system-level considerations.

Tech Stack: PyTorch, Python

Note: This project focuses on understanding training mechanics and model behavior rather than achieving state-of-the-art performance.

Learning Focus

Currently diving deep into:

  • ML systems architecture and design patterns
  • Production deployment strategies for ML models
  • Monitoring and maintaining ML systems in production
  • Cost optimization for inference at scale

Updated: February 2026