KMDS Update: Lean, Clean, and Practical ML Workflows

It’s been a while since my last post — I’ve been heads‑down refining KMDS and Tseda. Both tools have matured significantly, and I want to share where they stand today.

Why KMDS Matters

KMDS is now a modular, reproducible framework for ML workflows. It’s not about flashy demos — it’s about documented outcomes and a process that captures problems rigorously before solving them.

Three Problems We’ve Tackled

SBA Loans → predicting loan approvals based on custom feature engineering and an imbalanced learning workflow
Olist Retail → capturing temporal shopping preferences of Olist customers in Sau Paulo in 2017.
ITSM Ticket Resolution Time → using survival analysis to capture time to resolution of help desk tickets.

Each case shows KMDS delivering clarity, reproducibility, and measurable impact.

The Human Data Scientist Role

Assistants execute tasks, but the human expert defines context and reviews iterations. This loop ensures solutions are practical, context‑aware, and aligned with organizational priorities.

And here’s the critical point: good prompts, conceptual framing, precision, and interpretation of results are not generic skills — they are ML expert skills. Iterating and interrogating an AI assistant in a way that is meaningful to your use case requires deep understanding of machine learning. Without that expertise, the assistant cannot be guided to produce outcomes of real value.

Lean and Clean Positioning

KMDS isn’t about open‑ended demos. It’s about a structured consulting process: - Capture the problem with questionnaires
- Define workflows and units of analysis
- Solve with reproducible, documented outcomes

That’s how we deliver value — not by showing tools, but by solving problems with rigor and expert‑driven context.

Stay tuned for more demos and case studies. In the meantime, explore the repos and articles:

Citation

BibTeX citation:

@online{sambasivan2026,
  author = {Sambasivan, Rajiv},
  title = {KMDS {Update:} {Lean,} {Clean,} and {Practical} {ML}
    {Workflows}},
  date = {2026-07-25},
  url = {https://rajivsam.github.io/r2ds-blog/posts/kmds_changes/},
  langid = {en}
}

For attribution, please cite this work as:

Sambasivan, Rajiv. 2026. “KMDS Update: Lean, Clean, and Practical ML Workflows.” July 25. https://rajivsam.github.io/r2ds-blog/posts/kmds_changes/.