Hello, I'm

Rajat Patra

Machine Learning Engineer

Building AI-powered systems and seamless web platforms has always felt less like a job and more like a craft to me. The kind where you are never really done learning, and somehow that never gets old.

Lately that's Envault, a zero-knowledge secrets vault that's live in production, and a fantasy cricket XI predictor that won FIFS Gameathon 2.0 against 40+ Tier-1 teams. I obsess over the details most people never notice but always feel; that's just how I work.

Based in Bhubaneswar, India. You can reach me by email, or find me at @rajatpatra_.

Connect

If something you are building deserves that kind of attention, it would be great to connect. You can find me on any of the platforms below or reach out directly via email. Let's build something great.

Github Heatmap

One of the few things that shows my consistency.

Global Footprints

Community Impact.

xevrion/gyotakufix(app): the search window opens on the monitor being worked on

xevrion/gyotakuThe search window always opens on the primary monitor

vinzdg/codenotchWindows: carry the notch by six dots beside the settings button

#390

Where I've Worked

I showed up, I learned, I delivered.

GDG on Campus ITER
Bhubaneshwar, Odisha, India
Core Member
Part-Time
September 2024 - August 2025
- Organised and managed the backend operations for Techwiz 2024, an event presented by Coding Ninjas and GDG ITER, attended by 200+ students.
- Served as a key anchor for the Google Solution Challenge 2025, along with Google Tech Winter Break
AnchorEvent Management

Ekaayam
Bhubaneshwar, Odisha, India
Design Lead & Freelance Developer
Freelance
June 2024 - August 2025
- Engineered a ticketing platform that reduced manual booking effort by 70%.
- Redesigned and optimized the website to align with brand identity, leading to a 79% increase in website clicks.
Node.jsReactFlaskFirebaseRazorpay

Virtual Showreel
Bhubaneshwar, Odisha, India
Lead Editor
Full-Time
August 2024 - January 2025
- Managed the in-house production of Virtual Showreel, worked with the variation and content published for Instagram.
Video EditingContent Creation
Video Editor
Full-Time
May 2023 - January 2025
- Worked on Instagram short videos for official events hosted by Siksha 'O' Anusandhan.
- Secured leads for freelancing work as a Video Editor.
Video EditingInstagram Shorts

Tech Stack

Tools I reach for without thinking twice.

Language

Data & ML

Frontend

Backend & Database

Infrastructure

Workflow

Design

What I've Built

Things I actually built and shipped.

AES-256-GCM envelope encryption over a master/data key hierarchy with automatic rotation.

Redis-cached decryption, Supabase RLS, and immutable audit logs.

Sync CLI that blocks git-tracked files and provisions JIT GitHub access.

Next.js · Supabase · Redis · Tailwind

Envault

Live

Zero-knowledge secrets vault, running at envault.tech.

AES-256-GCM envelope encryption over a master/data key hierarchy with automatic rotation.

Redis-cached decryption, Supabase RLS, and immutable audit logs.

Sync CLI that blocks git-tracked files and provisions JIT GitHub access.

Next.js · Supabase · Redis · Tailwind

Drafts an optimal IPL fantasy XI inside the 100-credit cap and squad rules.

Gradient-boosted regressor for player output, PuLP for the constraint solve.

Trained on ball-by-ball data from Cricsheet, Kaggle and ESPNcricinfo.

Python · XGBoost · PuLP · Docker

Cricket Fantasy XI Predictor

1st · FIFS Gameathon 2.0

Won ₹10 lakhs against 40+ Tier-1 teams.

Drafts an optimal IPL fantasy XI inside the 100-credit cap and squad rules.

Gradient-boosted regressor for player output, PuLP for the constraint solve.

Trained on ball-by-ball data from Cricsheet, Kaggle and ESPNcricinfo.

Python · XGBoost · PuLP · Docker

ViT-B/16 fine-tuned on CIFAKE — 100k+ images, 98.2% F1 on the standard test set.

Held its accuracy against adversarial examples.

CUDA-accelerated training cut run time by 85%.

PyTorch · Torchvision · scikit-learn

AI Generated Image Classification

Tells real photographs from AI-generated ones.

ViT-B/16 fine-tuned on CIFAKE — 100k+ images, 98.2% F1 on the standard test set.

Held its accuracy against adversarial examples.

CUDA-accelerated training cut run time by 85%.

PyTorch · Torchvision · scikit-learn

CNNs, DenseNet, VGG16 and vision transformers compared at ~90% average accuracy.

Hyperparameter tuning and augmentation run across every architecture.

Scored on confusion matrices, F1, recall, precision and AUC-ROC.

Keras · TensorFlow · scikit-learn

Brain Tumor Detection

A model comparison study for flagging tumors in MRI scans.

CNNs, DenseNet, VGG16 and vision transformers compared at ~90% average accuracy.

Hyperparameter tuning and augmentation run across every architecture.

Scored on confusion matrices, F1, recall, precision and AUC-ROC.

Keras · TensorFlow · scikit-learn

Interface designed in Figma and Photoshop, then built out in Android Studio.

Python RFID reader publishing live facility status.

Google Maps API to surface the nearest hospitals to a user.

Python · Figma · Android Studio

HealSync

Tracks hospital facility availability in real time over RFID.