CipherLens — Classical Cipher Identification

Paste ciphertext. Get the cipher. No keys, no plaintext needed.

A machine learning web app that identifies 22 classical cipher types from raw ciphertext using statistical feature extraction and three complementary ML models.

Built as a B.Tech Project at IIIT Delhi, 2025–2026.

---

Features

• 22 cipher types across 6 families — monoalphabetic, polyalphabetic, transposition, polygraphic, fractionating, and modern block
• 3 model engines — Hybrid CNN (best accuracy), DL CNN (fastest), XGBoost (interpretable)
• 15 statistical features extracted per ciphertext — entropy, IoC, Kasiski analysis, bigram entropy, and more
• Feature importance — see which statistical properties drove each prediction
• Cipher encyclopedia — descriptions, formulas, and examples for all 22 ciphers

Models

Model	Architecture	Val Accuracy	Best For
Hybrid CNN	Character CNN + Statistical MLP (dual-input)	79.24%	Best overall — default
CNN Deep Learning	Character-level 1D CNN	68.47%	Fast, no feature engineering
XGBoost	Two-stage family → cipher with soft-routing	—	Interpretable, near-instant

Trained on 550,000 samples (cipher_MASTER_FULL_V4, 25k/cipher, MIN_LEN=100) with 15 statistical features. Full evaluation in docs/FINDINGS.md.

Supported Ciphers

Family	Ciphers	Count
Monoalphabetic Substitution	Caesar, Affine, Atbash	3
Polyalphabetic Substitution	Vigenere, Autokey, Beaufort, Porta	4
Transposition	Columnar Transposition	1
Polygraphic Substitution	Playfair, Hill, Four-Square	3
Fractionating	Bifid, Trifid, ADFGX, ADFGVX, Nihilist, Polybius	6
Modern Block	TEA, XTEA, Lucifer, LOKI, MISTY1	5
Total		22

Tech Stack

Layer	Technology
Frontend	Next.js 16, React 19, TypeScript, Tailwind CSS, shadcn/ui, Framer Motion
State	Zustand
Backend	FastAPI, PyTorch, XGBoost, scikit-learn
Infra	Docker Compose, Gunicorn/Uvicorn

Quick Start

With Docker (recommended)

git clone https://github.com/LordAizen1/cipherlens.git
cd cipherlens
docker compose up -d --build

• Frontend: http://localhost:3000
• Backend API + docs: http://localhost:8000/api/docs

Note: Trained model files are not included in the repo. Place them in backend/app/models/ before building. See Training to retrain from scratch.

Without Docker

# Backend
cd backend
python -m venv venv && source venv/bin/activate   # Windows: venv\Scripts\activate
pip install -r requirements.txt
uvicorn app.main:app --reload

# Frontend (new terminal)
cd frontend
npm install
npm run dev

Project Structure

cipherlens/
├── frontend/
│   └── src/
│       ├── app/          — Pages: Home, /ciphers, /about
│       ├── components/   — UI components
│       ├── hooks/        — Zustand store
│       └── lib/          — API client, types, cipher constants
├── backend/
│   ├── app/
│   │   ├── routers/      — /api/predict, /api/health
│   │   ├── services/     — Feature extraction + 3 inference engines
│   │   └── models/       — Trained .pkl and .pth files (not in git)
│   └── scripts/          — Dataset generation + training scripts
├── data/                 — Dataset CSV (not in git)
├── docs/
│   ├── FINDINGS.md       — Full model evaluation report
│   └── Project_Report.md — BTP project report
├── docker-compose.yml
└── CONTRIBUTING.md

Training

Models were trained on the IIIT Delhi Precision cluster (NVIDIA H100 MIG GPU).

# 1. Generate dataset (cluster recommended — ~2–3 hrs locally)
python backend/scripts/generate_dataset_v4.py

# 2. Train XGBoost (~2 min, CPU)
cd backend && python scripts/train.py

# 3. Train DL CNN (~5 min, GPU)
python scripts/train_dl.py

# 4. Train Hybrid CNN (~26 min, GPU)
python scripts/train_hybrid.py

SLURM job scripts (train_job.sh, train_dl_job.sh, train_hybrid_job.sh) are included for cluster submission.

Contributing

Contributions are welcome — new ciphers, better models, UI improvements. See CONTRIBUTING.md for guidelines.

Team

Name	Roll	GitHub	LinkedIn
Dhruv Verma	2022172	@dhruv22172	LinkedIn
Maulik Mahey	2022282	@maulik-dot	LinkedIn
Md Kaif	2022289	@LordAizen1	LinkedIn
Sweta Snigdha	2022527	@cypherei00	LinkedIn

Supervised by Dr. Ravi Anand — IIIT Delhi

License

MIT