[ M.Tech Student  ·  ML Researcher  ·  Builder ]

Ayush
Ranjan

Integrated M.Tech in Computational & Data Science at VIT Bhopal. I build ML systems, work on NLP problems, and have published research in cryptography. I care more about whether to trust a model than whether it scores well on paper.

See the Work Resume ↗
scroll
// chapter_01.about

Building with Data,
Thinking with Doubt

I'm an Integrated M.Tech student at VIT Bhopal (Computational & Data Science). My work covers NLP, ML forecasting, LLM post-training, and cryptography research — all connected by one question: can we actually trust what this model is saying?

Interned at Ethara.ai as an LLM Post Trainer, hold a Google IT Support certification, and published research in a Q1 Scopus journal. On Codeforces as Raj_ay (Pupil) — 300+ DSA problems on LeetCode.

0
DSA Problems Solved
0
Public ML Projects
0
Scopus Paper Published
Pupil
Codeforces Rank · Raj_ay
ayush_ranjan.json
{
  "role": "ML Engineer & Researcher",
  "institution": "VIT Bhopal · M.Tech CDS",
  "status": "open_to_opportunities",
  "papers": 1 // Q1 Scopus · 2026
  "focus": ["NLP", "LLM alignment", "model reliability"]
}
◈ Published Researcher ◈ LLM Post Trainer · Ethara.ai ◈ Google IT Certified ◈ Codeforces Pupil · Raj_ay ◈ 300+ DSA · LeetCode ◈ Cryptography · GF(3) LFSR
AR ML · NLP · CRYPTO
// chapter_02.skills

The Toolkit

🧠
Machine Learning
PyTorch TensorFlow Scikit-learn XGBoost Random Forest
💬
NLP
Transformers BERT TF-IDF LSTM spaCy NLTK
📊
Data Science
Pandas NumPy Matplotlib Seaborn Plotly
🔐
Cryptography
LFSR GF(3) Fields Image Encryption Chaos Theory
⚙️
Engineering
Python TypeScript FastAPI SQL Git Docker
🤖
LLM / AI Ops
LLM Fine-tuning RLHF Post-training Prompt Eng.
// chapter_03.projects

Things I Built

Full Projects + Demos →
AI Infrastructure · Multi-Agent
NoteMind FEATURED
Graph-based persistent memory for multi-agent AI swarms. Each agent writes atomic notes into ChromaDB, visible to every other agent via semantic search. Token cost stays O(k) instead of exploding O(n²).
Gemini 2.5 ChromaDB Next.js FastAPI
View Project →
Chrome Extension · Gemini Vision
MangaLens SHIPPED
Chrome extension that translates text inside manga and manhwa images in-place using Gemini Vision. Detects speech bubbles, fills background, overlays translation on the canvas. v1.0.8, live.
JavaScript Gemini Vision Canvas API Released ●
View Project →
ML · Finance · Reliability
MarketLens
CNN + Transformer for probabilistic stock forecasting. P10/P50/P90 bands. The RAPID framework tells you when not to trust the prediction — built in, not bolted on.
PyTorch Transformers RAPID
View Project →
NLP · Sentence Embeddings
AI Resume Analyzer
Scores a resume against a job description using TF-IDF + sentence embeddings (all-MiniLM-L6-v2) + skill Jaccard overlap. Shows match score and skill gaps. Streamlit app with OpenVINO inference.
sentence-transformers TF-IDF Streamlit
View Project →
NLP · Legal Tech
Legal NLP Classifier
93% accuracy classifying 5,000+ Supreme Court judgments. TF-IDF + SVM with keyword evidence highlighting — the model shows its reasoning.
TF-IDF SVM 93% Acc.
View Project →
Computer Vision · Deployed
TeachAI
Face recognition + GPS geofencing for smart attendance. Present only when face matches AND device is within classroom radius. Live on Vercel.
Face Recognition TypeScript Live ●
Live App → ↗ vercel
Research · Published
Ternary LFSR Cipher
Q1 Scopus paper: GF(3) LFSR stream cipher for image encryption. NPCR 98.04%, entropy 6.80 bits. Interactive register visualizer on the research page.
GF(3) LFSR Published ●
Read Paper →
"Should we even trust this prediction?"
— the question that drives every model I build
// chapter_04.experience

The Path So Far

May 2026 · Research
Research Paper Published
Advances in Decision Sciences · Q1 Scopus · CiteScore 8.3
"A New Approach for Image Security Enhancement Using Ternary Logic LFSR for Cryptographic Applications" — Vol 30(3), 2026, pp. 184–214. Novel GF(3) feedback mechanism achieving 98.04% NPCR and 6.80-bit entropy for healthcare image security.
GF(3) LFSR Image Encryption Q1 Scopus
Feb 2026 – Apr 2026 · Internship
LLM Post Trainer Intern
Ethara.ai · Remote · Completed
Trained and evaluated LLM outputs via RLHF. Curated fine-tuning datasets, wrote adversarial prompts, scored responses for HHH alignment (Helpfulness, Honesty, Harmlessness). Contributed to post-training pipelines at the model feedback stage.
LLM RLHF Post-Training HHH Alignment
Feb 2026 · Certification
Google IT Support Professional
Google Career Certificates / Coursera · Issued 23 Feb 2026
6-course certification covering networking, operating systems, system administration, security fundamentals, and technical support. Verified credential on Credly.
Networking OS Security Google
2023 → Present
Integrated M.Tech CSE (Computational & Data Science)
Vellore Institute of Technology, Bhopal
5-year integrated program. ML theory, algorithms, data engineering, systems design. Active competitive programmer — Pupil on Codeforces (Raj_ay), 300+ DSA problems solved.
M.Tech CSE Data Science Algorithms
// chapter_05.research

Published Research

Advances in Decision Sciences · Q1 Scopus · CiteScore 8.3 · DOI: 10.47654/v30y2026i3p184-214
A New Approach for Image Security Enhancement Using
Ternary Logic Linear Feedback Shift Register
for Cryptographic Applications
Proposed a novel ternary-field (GF(3)) LFSR-based stream cipher for encrypting medical and sensitive images. Unlike binary LFSRs, the base-3 feedback mechanism exponentially expands the keyspace, resisting brute-force and statistical attacks. Validated across entropy, NPCR, UACI, and pixel-correlation metrics.
0%
NPCR (Pixel Change Rate)
0
Shannon Entropy
0
Journal CiteScore
Interactive Deep Dive → Original Paper ↗
// chapter_06.competitive

Grinding the Fundamentals

LeetCode
0+
Problems Solved
Codeforces
Pupil
Rank — handle: Raj_ay
Focus Areas
DP · Graphs · Trees
Primary Topics
// fin.contact

Let's Build
Something Real

Whether you're working on an ML problem, need someone who asks hard questions about model reliability, or just want to talk cryptography — reach out.

Email GitHub LinkedIn LeetCode Codeforces Resume