Welcome

Senior Machine Learning Engineer working at the seam of data engineering and machine learning — most days that means enterprise RAG, secure deployments, and the unglamorous data plumbing that decides whether any of it actually helps anyone.

Recent writing

• Lessons learned using Arize Phoenix
  May 15, 2026

  LLM observability is a really fast moving field and there’s a lot of tools out there masquerading as the state of the art LLM observability tool.

• Claude code reigns supreme
  Apr 6, 2026

  ...and it's not close, not even a little bit. I spent money for all the subscription plans so you don't have to. Don't be me. Just get Claude Code Pro.

• The genius Apache stepchild, Presto
  Apr 6, 2026

  an under-rated essential component of any cloud service ecosystem

• Integrity - invisible force for good
  Jan 16, 2026

  In high school, there’s all sorts of different conceptual lessons that teachers impart on students.

• Takeaways from the AWS ReInvent 2023 Conference
  Jul 24, 2024

  I totally forgot how fun conferences are!

Selected projects

• amadabhu.github.io

  shipped

  This site, a Jekyll-flavored personal site with a small Python build pipeline so it renders without Ruby.

  • Jekyll
  • Liquid
  • Python
  • Markdown
  • GitHub Pages
  • GitHub ↗

  The site you're reading. Templates are Jekyll-flavored Liquid + Markdown, but the build runs through a small Python pipeline (scripts/build_site.py) so I can iterate locally without installing Ruby. GitHub Actions runs the same Python build on push and deploys the artifact to Pages.

• WIFX

  shipped

  Women's International Football Rankings — analytics dashboard live at wifxfootball.com.

  • Python
  • scikit-learn
  • XGBoost
  • PyTorch
  • Hugging Face Spaces
  • GitHub Actions
  • Live site ↗

  A composite-metric ranking system for women's football across players, club teams, national teams, and confederations.

  Match predictions are produced by :
  - Poisson goal models
  - Gradient-boosted trees
  - Small neural nets
  - Ensemble models
  

  Then, selected by 5-fold time-series CV log-loss high-confidence calls require agreement across a 12-model consensus panel split into distinct model families.

• notebook2ppt

  public

  Convert a Jupyter notebook into a PowerPoint deck — headings become slides, figures and tables flow into the layout.

  • Python
  • nbformat
  • python-pptx
  • Pillow
  • pytest
  • GitHub ↗

  nbconvert --to slides produces reveal.js HTML, which is great in the browser and useless when somebody on the other side of the org wants a .pptx they can edit. notebook2ppt fills the gap: read the notebook you already wrote, write a real PowerPoint file.

  The mapping is heading-driven — # and ## markdown headings open slides; markdown bodies, matplotlib PNGs, pandas HTML tables, and stream outputs become typed Blocks placed by a python-pptx renderer. A two-stage pipeline (notebook → IR → .pptx) keeps extraction independent of layout, so swapping in a different render target later (Google Slides, reveal.js, plain HTML) doesn't require redoing the parsing work.

  Ships with a CLI (notebook2ppt analysis.ipynb -o deck.pptx), a Python library, three example notebooks with real baked-in outputs, and a small pytest suite.

Full list on the projects page.

Contact

• GitHub — public code, forks, and a deliberately messy scratchpad.
• Substack — irregular notes on whatever rabbit hole has me captive.
• Email me : first.last at [gmail.com,protonmail.com]