Hari Baskar
Reliability AI & Industrial Systems
I turn real-world signals into reliable decisions — and reliable decisions into products.
I'm a data scientist and reliability practitioner. I work on the chain from sensor to signal to model to decision to maintenance action — and I write about what actually breaks along the way.
Read my writing Subscribe to The Field Notes
Senior data scientist in wind energy · Asset Reliability Practitioner & Certified Vibration Analyst Cat II (Mobius Institute) · ~10 years of ML on real-world industrial data
Start here
If you're new, this is what I'm about: industrial AI doesn't fail at the model — it fails at the decision. A prediction that nobody trusts, an alarm that fires too often, a lead time that arrives too late to act on. I've spent a decade watching where signal turns into (or fails to turn into) action, and that's what I write about.
Over the next decade I want to understand the whole reliability stack — from physical failure modes and sensing, through models and thresholds, to edge systems and hardware. This site is the working record.
Writing
Essays and field notes on predictive maintenance, condition monitoring, industrial ML, and the slow move from software toward sensors and hardware.
- From spectra to decisions — condition monitoring as a decision system, not a dashboard. New.
- What changes when an ML model touches the field — why a good model can still make a bad maintenance system.
- Model accuracy is not maintenance value — false alarms as an organizational trust tax.
Coming next:
- What a vibration analyst knows that a data scientist doesn't — physics before features.
- The case for boring models — why simple, inspectable models survive the field.
- Moving down the stack — why the next decade of reliability lives at the edge, next to the machine.
Want them in your inbox? That's what The Field Notes is for.
Selected projects
- Anomaly detection on streaming data
HBOS, denoising autoencoders, and decision trees to catch operational outliers before they became incidents. My first real lesson in the cost of false alarms. - Model interpretation in practice
SHAP, LIME, and ELI5 to explain individual predictions. If an engineer can't see why a model flagged something, they won't act on it. - Churn & performance prediction with survival analysis
Survival models, HMMs, and gradient boosting to predict decline before it happened. Same math that underpins remaining-useful-life work. - Invoice extraction without templates
Tesseract OCR plus clustering on bounding boxes. A reminder that most real-world data problems are messy-input problems. - Handwritten Tamil character recognition
OpenCV + k-NN, 90%+ accuracy, back in 2016. Kept here because it's where I learned that signal preprocessing decides everything downstream.
See all 13 projects, from a 2014 code editor to industrial anomaly detection →
Now
September 2026.
Learning: deepening vibration analysis beyond my Cat II — sensor placement and what it makes observable.
Building: relaunching this site as my working notebook, and getting a first public reliability notebook out on open data.
Exploring: where reliability intelligence meets edge hardware — the start of a ten-year thread.
Updated monthly-ish. Last update: September 2026.
About
I'm Hari Baskar, a senior data scientist and reliability practitioner based in India. I've spent about ten years turning noisy real-world data into decisions people can act on — first in NLP and computer vision, then in wind energy, where the data comes off physical machines and being wrong has a maintenance bill attached.
Along the way I picked up credentials most data scientists don't bother with — Asset Reliability Practitioner and Certified Vibration Analyst Category II from the Mobius Institute — because I think you can't model a machine you don't understand.
The full story: every role, internship, certification, award, and a couple of security disclosures →
Contact
The best way to reach me is email: haribaskar98@gmail.com. I read everything, especially notes from reliability, maintenance, and industrial-ML folks.