Pradosh Pattanayak

Senior Technical Architect

[Pradosh Pattanayak @er_pradosh]

@castledmallard I hate right wing conservatives.
And whenever I see a leftist, want pigeon to shit on them for PEACE.

[Pradosh Pattanayak @er_pradosh]

I hate right wing conservatives.
And whenever I see a leftist, want pigeon to shit on them for peace.

[Pradosh Pattanayak @er_pradosh]

The Pride of India.
Scientific name: Lagerstroemia speciosa
Local Name: Jarula (ଜରୁଳ)
What’s it called in your local language?

#PrideOfIndia #NatureIndia #JarulTree #IndianFlora

[Pradosh Pattanayak @er_pradosh]

Built a regression model from scratch - no libraries, just math, matrix ops & gradient descent.

From data leakage to production-ready design, this is how ML really works under the hood.

Final prediction: 295.92 🚀

Read: https://t.co/14qJ2SLqOo
#MachineLearning #DeepLearning

[Pradosh Pattanayak @er_pradosh]

Data normalization isn’t optional - it’s what keeps gradients from exploding 🚀

Z-score standardization brings features to the same scale, stabilizing learning & speeding convergence.

Read👇
https://t.co/doHQcY2CxR
#AI #MachineLearning #DeepLearning #DataScience #Math #Python

[Pradosh Pattanayak @er_pradosh]

Blew up my gradient descent in 38 epochs 😅
Weights → 1e+150, Loss → 1e+300 → OverflowError
Lesson: feature scaling isn’t optional — it’s survival.
Unscaled inputs = exponential explosion.

Read 👇
https://t.co/n9YAFPQKOk
#MachineLearning #AI #GradientDescent #Math #Debugging

[Pradosh Pattanayak @er_pradosh]

Watching gradient descent actually learn 🔥

From random weights → near-perfect fit in ~20 epochs.
Partial derivatives isolate the blame: weight adjusts slope, bias shifts the line.

https://t.co/DH6SFOMkpF

#MachineLearning #AI #GradientDescent #Math #LearningInPublic

[Pradosh Pattanayak @er_pradosh]

Backprop finally clicked 🔥

As error → 0, gradient → 0
So updates become tiny → learning slows down
Big steps when wrong, tiny steps near correct = asymptotic convergence

🔗 https://t.co/f2Nx9Gzul1

#AI #ML #DeepLearning #Backpropagation #GradientDescent

[Pradosh Pattanayak @er_pradosh]

Derivative gave direction.
Tangent revealed slope.
Gradient Descent made it actionable.
+ve slope → move left
-ve slope → move right

From intuition → math → code
https://t.co/1xeu69KU8W

#MachineLearning #AI #GradientDescent #Math #LearningInPublic

[Pradosh Pattanayak @er_pradosh]

Built a multi-output ML model from scratch

Biggest realizations:

• Matrix multiplication = multiple mappings
• Transpose = alignment, not a trick
• Errors > tutorials
I didn’t learn ML by reading. I learned it by breaking it.
Full journey ↓
https://t.co/8Idn6MChIo

[Pradosh Pattanayak @er_pradosh]

@Valenti78116865 Thanks for sharing. I will check it. 👍🏼

[Pradosh Pattanayak @er_pradosh]

Small bug. Big learning.

Wrong: https://t.co/A2SI2h0lIg_vector(w) + b
👉Right: add bias per prediction

Built this in my regressor from scratch →
https://t.co/F1Bx3B84SV

#MachineLearning #AI #BuildInPublic

[Pradosh Pattanayak @er_pradosh]

@ManOfFocus_ Learn . Build . Peace

[Pradosh Pattanayak @er_pradosh]

@Memelords212 Clicked today

[Pradosh Pattanayak @er_pradosh]

@cricket_badger It happens in Soccer as well. During a period a team dominates the sports. In this era in Cricket, it’s India.

[Pradosh Pattanayak @er_pradosh]

@Gaurav_7887 14 Years+ and running

[Pradosh Pattanayak @er_pradosh]

@Jigspunadiya f(you) = ⊥

[Pradosh Pattanayak @er_pradosh]

@relatablehuyrrr Cold Lassi

[Pradosh Pattanayak @er_pradosh]

@Amelia558rs I vouch it’s 69.

[Pradosh Pattanayak @er_pradosh]

@Phenom962 Ye dekh ke Usman Tariq k liye izzat badh gaya hai 😂