Seerror / Datasets
Training Data

Seerror Datasets

India-specific AI training data for cybersecurity, privacy and digital safety, synthetic, auditable, and built for real-world use.

Synthetic No real PII
India First
Audited Reproducible
Available now
Cyber-safety Free for India June 2026

Seerie Dataset

4,180 India-specific cybersecurity conversations covering UPI fraud, OTP scams, Aadhaar misuse, digital arrest, sextortion and more. English, Hinglish and Hindi. White paper, dataset paper and reproducible audit included.

4,180 records 3 languages Instruction tuning
Open models
Current · v4 Apache 2.0 Hugging Face

Seerie v4 · v407

Current offline India fraud AI. Fine tuned from Qwen3 1.7B on 140,813 conversations. 86/100 safe on the previous test suite. GGUF for phones, LM Studio, Ollama, and llama.cpp.

1.7B params ~1.1 GB Q4 Apache 2.0
Previous · v3 Apache 2.0 Hugging Face

Seerie v3 · v349

Previous release. 107,575 conversations. 87/100 safe on a hand rating. Still available for comparison.

1.7B params ~1.1 GB Q4 Apache 2.0
Previous · v2 Apache 2.0 Hugging Face

Seerie v2 · v224

Previous release. 36,669 conversations across 22 core fraud types. Still available for comparison and older evaluations.

1.7B params ~1.1 GB Q4 298 downloads
Coming soon
Pipeline · 2026

More datasets in development

Additional India-specific training sets for phishing detection, privacy compliance, and on-device security assistants are in the pipeline. Follow us for announcements.

Building with our data or need a custom dataset for your use case? Reach out with a short note about what you're working on.

Contact the team