An offline assistant for Indian digital fraud
Seerie is a compact, on-device language model that helps people in India recognize, respond to, and report digital fraud, including UPI scams, phishing, OTP and KYC fraud, digital arrest scams, and more. It runs fully offline, so user questions and personal data never leave the device.
This repository provides GGUF quantizations of Seerie for llama.cpp, Ollama, and LM Studio. All files are built from the v2 model, trained on the v224 dataset. Full-precision safetensors live at seerror-technologies/seerie. The Seerie app is live on Google Play. Install links also sit on the Downloads & install page.
India first safety knowledge
Trained to answer the questions people actually ask after a scare, in the language they use.
A real Hinglish exchange
Phir cybercrime.gov.in par complaint file karein ya 1930 helpline par call karein,
jitni jaldi report karenge, paise wapas milne ke chances utne zyada honge.
Apna transaction ID aur bank statement screenshot ready rakhein complaint ke liye.
36,669 conversations across India fraud types
Responses point people to official channels such as the 1930 National Cyber Crime Helpline or cybercrime.gov.in. Seerie answers in the language of the question.
| Topic | Conversations |
|---|---|
| OTP / KYC Fraud | 10,416 |
| UPI / Payment Fraud | 8,906 |
| Aadhaar Misuse | 4,671 |
| SIM Swap / Telecom Fraud | 2,324 |
| Phishing (SMS, Email, Call) | 2,102 |
| General Cyber Safety / Other | 1,904 |
| IT Act / Cyber Law Guidance | 1,056 |
| Digital Arrest Scam | 730 |
| Lottery / Prize Scam | 639 |
| Investment / Trading Scam | 638 |
| Ransomware / Malware | 612 |
| Loan App Fraud | 585 |
| Sextortion / Blackmail | 486 |
| Deepfake / AI Voice Cloning Scam | 384 |
| Social Media Account Hacking | 329 |
| OLX / Marketplace Scam | 279 |
| Matrimonial / Romance Scam | 193 |
| Electricity / Utility Bill Scam | 141 |
| Data Privacy / DPDP Act | 116 |
| Job / Work From Home Scam | 66 |
| Cyberbullying / Harassment | 48 |
| Army / Defence Impersonation Scam | 44 |
| Total | 36,669 |
Fine tuned from Qwen3 1.7B
Seerie is a causal language model, instruction-tuned for conversation. It is fine-tuned from Qwen/Qwen3-1.7B (Apache 2.0) using QLoRA with Unsloth, 3 epochs, on 36,669 curated conversations (dataset v224). The companion instruction set is documented on the Seerie dataset page.
Available GGUF files
| File | Size | Notes |
|---|---|---|
seerie-q4_k_m.gguf Recommended |
1.11 GB | Default. Phones and low-memory devices. Seerie app Brain download. |
seerie-q6_k.gguf |
1.42 GB | Higher-end phones and laptops. |
seerie-f16.gguf |
3.45 GB | GPU machines, evaluation, or further quantization. |
Get the files from huggingface.co/seerror-technologies/seerie-GGUF.
LM Studio, Ollama, llama.cpp
Ollama
ollama run hf.co/seerror-technologies/seerie-GGUF:Q4_K_M
llama.cpp
llama-cli -hf seerror-technologies/seerie-GGUF:Q4_K_M
Or with a local file:
llama-cli -m seerie-q4_k_m.gguf -p "Mera UPI se paise kat gaye, kya karoon?"
LM Studio
- Search for
seerror-technologies/seerie-GGUF, or download a.gguffile from Hugging Face - Load the model and leave the system prompt empty
- Use the chat template embedded in the GGUF file
- Set temperature to 0.3 to 0.4
Recommended settings
| Setting | Value |
|---|---|
| Temperature | 0.3 to 0.4 |
| System prompt | Leave empty (behaviour is trained into the model) |
| Chat template | Use the template embedded in the GGUF file |
63 / 70 passed (90%)
Seerie v2 was evaluated on an internal benchmark of 70 test cases across 14 categories, including UPI fraud, digital arrest scams, OTP handling, child safety, legal, medical and financial guidance boundaries, and privacy claims. Overall score: 63/70 passed (90%).
What changed in v2
v2 (September 2026). Retrained on dataset v224: 36,669 conversations, up from about 25,000. Full 3-epoch run. Added 9 Indian languages in native scripts, up from English, Hindi, and Hinglish. Expanded coverage to deepfake and AI voice cloning, matrimonial and romance scams, ransomware, and more. Rebuilt all quantizations; added Q6_K and F16.
v1. Initial public release, trained on about 25,000 conversations in English, Hindi, and Hinglish. Q4_K_M quantization.
What Seerie is not
- Not a substitute for filing an official police or cybercrime complaint. Always direct urgent cases to 1930 or cybercrime.gov.in.
- Knowledge is limited to patterns in the training data. New scam types after training may not be recognized.
- As a 1.7B-parameter model, it is best suited to quick, practical guidance rather than detailed legal analysis.
- Like all language models, it can produce incorrect or incomplete information. Verify important details with official sources.
- Not a government product and not binding legal, financial, or medical advice. Do not use it to make automated decisions about people or transactions.
Where to go next
How to cite
@misc{seerie2026,
title = {Seerie: An On-Device Language Model for Digital Fraud Protection in India},
author = {{Seerror Technologies}},
year = {2026},
howpublished = {\url{https://huggingface.co/seerror-technologies/seerie-GGUF}}
}
@misc{tiwari2026seerie,
author = {Tiwari, Jay},
title = {Seerie: An Offline 1.7B-Parameter Assistant for Scam Prevention in India, and a Manual Audit of Its Failures},
year = {2026},
note = {Version 224},
publisher = {Zenodo},
doi = {10.5281/zenodo.23110076},
url = {https://doi.org/10.5281/zenodo.23110076}
}