Folio — EditionSakina Saifuddin© 2026
2021

A journey through code, data, and intelligent systems.

Folio — EditionSakina Saifuddin© 2026
SS
01Intro02Work03What I Build04Achievements05Experience06Expertise07Contact
Case studyClick a page, swipe, or use ← → to turn

PREDICTING POWERPLANT FAILURESWITH MACHINE LEARNING

POWER GUARD360

  • 2nd of 70+ teams at CodeRED Orion and a $500 Best Pitch prize.
  • Machine learning flags failing turbines and explains why.
  • Real sensors, a 3D plant, and a phone app for the engineer.
2ndof 70+ teams
96%precision on unseen data
Visit the portfolio
website at:sakina-saifuddin / works
19192
Power Guard 360Sakina Saifuddin

PREDICTING POWERPLANT FAILURESWITH MACHINE LEARNING

Keep the AI data center powered
Fig. 1 — Our model of the plant and the AI data center it feeds.1
  1. Task
  2. My work
  3. Results
  4. Outcome

Keep the AI data center powered

  • Gas turbines power an AI data center directly.
  • If a turbine fails without warning, the computers go dark.
  • The plant needs to catch failures early and fix them fast.
24 hrsto build it at CodeRED Orion
2
Two models that explain themselves
Fig. 2 — The model flags GT-03 and lists the readings behind it.3
  1. Task
  2. My work
  3. Results
  4. Outcome

Two models that explain themselves

  • Model 1 gives each turbine a calibrated failure risk.
  • Model 2 names the failure type: heat, power, overstrain or tool wear.
  • SHAP lists the three readings that drove each alert.

Tools: Python · scikit-learn · SHAP · FastAPI

4
Real sensors drive one turbine
Fig. 3 — The live sensor board next to the plant view.5
  1. Task
  2. My work
  3. Results
  4. Outcome

Real sensors drive one turbine

  • A spinning motor, a barometer and an air quality sensor feed turbine GT-08.
  • The board posts readings twice a second.
  • Low air quality at gas metering raises a gas leak alert.

Tools: Arduino UNO R4 WiFi · Node.js · WebSockets

6
The whole plant in 3D
Fig. 4 — The 3D digital twin with live readings for every turbine.7
  1. Task
  2. My work
  3. Results
  4. Outcome

The whole plant in 3D

  • Built the 3D plant model with a Blender script.
  • Click any turbine to see its readings and the model's diagnosis.
  • Designed the version 2 interface, logo and phone layout.

Tools: Blender · React · three.js · TypeScript

8
From alert to engineer
Fig. 5 — The engineer's phone gets the job.9
  1. Task
  2. My work
  3. Results
  4. Outcome

From alert to engineer

  • The manager approves a repair and picks an engineer.
  • The engineer's phone buzzes with the job.
  • The control room watches them walk to the turbine on the 3D map.

Tools: Expo · React Native

10
Accurate on data it never saw
Fig. 6 — A possible gas leak raised from the live air quality sensor.11
  1. Task
  2. My work
  3. Results
  4. Outcome

Accurate on data it never saw

Tested on 2,000 readings held back from training.

96%precision
78%recall
0.90PR-AUC (XGBoost 0.88)
0.93macro F1 for failure type
12
2nd place out of 70+ teams
Fig. 7 — Our team on stage with the $500 Best Pitch/Demo check.13
  1. Task
  2. My work
  3. Results
  4. Outcome

2nd place out of 70+ teams

  • Led a team of undergraduates to 2nd place at CodeRED Orion.
  • Won $500 for Best Pitch/Demo.
  • Still live at powerguard360.us.
2ndof 70+ teams
$500cash prize
14
15

Technologies

scikit-learn + SHAP
Predicts failures and explains them.
FastAPI
Serves the models to the plant server.
Node.js + WebSockets
Runs the plant and pushes live data.
React + three.js
The 3D plant and the dashboard.
Expo
The engineer's phone app.
Arduino
Live turbine and gas leak sensors.
What I did
  • Led a team of undergraduates and split the work by each person's strengths.
  • Built both machine learning models, the SHAP explanations and the model service.
  • Built the 3D plant model and the version 2 interface.
  • Added the public view with admin sign-in and deployed the plant server.

Seven of the eight turbines replay the public AI4I 2020 dataset; the sensor board drives the eighth.

Live websiteYouTube demoDevpostThe achievement
16

POWER GUARD360

  • 2nd of 70+ teams at CodeRED Orion and a $500 Best Pitch prize.
  • Machine learning flags failing turbines and explains why.
  • Real sensors, a 3D plant, and a phone app for the engineer.
2ndof 70+ teams
96%precision on unseen data
Visit the portfolio
website at:sakina-saifuddin / works
19192
Cover