I work on applied machine learning systems, with a focus on taking models from experimentation to reliable, real-world deployment. My interests sit at the intersection of production ML, deep learning, and generative AI, especially where system design and evaluation matter as much as model performance.

Most of my work explores how research ideas translate into scalable, maintainable ML systems beyond notebooks and demos.

• End-to-end machine learning systems used in production
• Deep learning models for vision, sequence, and ranking problems
• Production ML pipelines with deployment, monitoring, and iteration in mind
• Applied generative AI and LLM-based systems
• Data pipelines, evaluation workflows, and experimentation frameworks

I care a lot about ownership, clarity, and making ML systems work reliably over time.

Here are a few projects that represent my current interests and approach:

• Emotion Dynamics in Podcast Conversations
  Transformer-based ranking model that predicts podcast engagement using emotional trajectories and time-series analysis.
  https://github.com/anshulraj10/podcast-emotion-dynamics

• Multi-LLM Agent Feedback Loop
  A multi-agent LLM system where models critique and refine each other’s outputs to reduce hallucinations and improve reliability.
  https://github.com/anshulraj10/multi-llm-agents-feedback

• AI Music Generator
  Emotion-conditioned music generation pipeline combining NLP-based emotion detection with sequence modeling.
  https://github.com/anshulraj10/ai-music-generator

• Automated Author & Reference Extraction
  NLP pipeline to extract and normalize structured metadata from unstructured academic PDFs.
  https://github.com/anshulraj10/author-extraction

• Generative Face Aging App
  Computer vision application exploring generative image transformation with practical deployment considerations.
  https://github.com/anshulraj10/face-aging-app

I approach machine learning as a systems problem, not just a modeling task.

Some principles I try to follow:

• Prefer simple, well-evaluated solutions over complex but fragile ones
• Design ML pipelines with deployment, monitoring, and iteration in mind
• Treat data quality, evaluation, and failure modes as first-class concerns
• Translate research ideas into practical systems that teams can operate and trust

I enjoy working on problems where engineering decisions matter as much as algorithmic choices.

Machine Learning & AI

• PyTorch, TensorFlow, Scikit-learn
• Deep Learning, Computer Vision, NLP
• Transformers, LLM-based systems, Generative AI

Production ML & Systems

• Model deployment, CI/CD for ML, monitoring
• Docker, Jenkins, Linux
• Edge and real-time ML systems

Data & Programming

• Python, SQL
• Pandas, NumPy
• Experimentation and evaluation tooling

Some repositories are exploratory in nature. I often use GitHub to test ideas, prototype systems, and document learnings as I go. Not everything here is polished, but most projects reflect real problems I was curious to understand or solve.

• Website: https://anshulraj.com
• LinkedIn: https://www.linkedin.com/in/anshul-raj
• Email: anshul.raj2020@gmail.com

This README is intentionally kept simple and timeless. It reflects how I work, not just what I work with.