Downloads · 30 days
49
8% of all-time downloads
GOVINDFROM/aegis-gemma-4-e4b-it-scam-defense-ollama-gguf-v1
aegis-gemma-4-e4b-it-scam-defense-ollama-gguf-v1 is a text generation model from GOVINDFROM. Use it when you need the model to write or continue text. It is set up for gguf. The card lists the license as gemma.
AEGIS is a locally deployable Gemma 4 E4B instruction-tuned model package for scam-defense analysis. It is tuned for one job: analyze a phone-call transcript for epistemic manipulation and return structured JSON descr…
Downloads · 30 days
49
8% of all-time downloads
All-time downloads
583
Public
Repo size
5.3 GB
Likes
1
Public
Click a slice to open those files.
.gguf5.3 GB · 100%
From the Hugging Face model README
AEGIS is a locally deployable Gemma 4 E4B instruction-tuned model package for scam-defense analysis. It is tuned for one job: analyze a phone-call transcript for epistemic manipulation and return structured JSON describing how a scammer is trying to distort the target's decision-making process.
This repository is the Ollama / GGUF release package for:
The package is designed for the GOVINDFROM/aegis-gemma-4-e4b-it-scam-defense-ollama-gguf-v1 Hugging Face repo.
Given a conversation transcript, AEGIS scores six manipulation vectors:
belief_installationverification_suppressionurgency_fabricationauthority_hijackingemotional_floodingexit_path_closureIt returns:
0-100integrity_scorerisk_levelgemma-4-e4b-it.Q4_K_M.gguf: quantized GGUF model for local inferenceModelfile: Ollama model definition with the AEGIS system behavior and prompt templateREADME.md: usage and model cardgoogle/gemma-4-E4B-itGGUFQ4_K_MThis release is derived from a fine-tuned Gemma 4 E4B instruction model adapted for scam-defense transcript analysis.
162e-433964420480.4065NVIDIA A100-SXM4-80GBUse this model to:
Do not use this model as:
git lfs install
git clone https://huggingface.co/GOVINDFROM/aegis-gemma-4-e4b-it-scam-defense-ollama-gguf-v1
cd aegis-gemma-4-e4b-it-scam-defense-ollama-gguf-v1
ollama create aegis -f Modelfile
ollama run aegis
huggingface-cli download GOVINDFROM/aegis-gemma-4-e4b-it-scam-defense-ollama-gguf-v1 --local-dir .
ollama create aegis -f Modelfile
ollama run aegis
Paste a transcript like this into Ollama:
Analyze this phone call for epistemic manipulation:
Caller: Hello, this is Agent Williams from the IRS Criminal Investigation Division.
We found major discrepancies in your tax filings.
A federal arrest warrant has been issued in your name.
You must pay immediately via gift cards to stop enforcement.
Do not hang up or contact anyone else.
Expected behavior:
authority_hijackingurgency_fabricationverification_suppressionrisk_level near CRITICALintegrity_scoreThe model is configured to emit JSON in this structure:
{
"manipulation_vectors": {
"belief_installation": 95,
"verification_suppression": 90,
"urgency_fabrication": 100,
"authority_hijacking": 100,
"emotional_flooding": 85,
"exit_path_closure": 95
},
"integrity_score": 5,
"risk_level": "CRITICAL",
"explanation": "The caller is using false authority, panic, and isolation tactics to block independent verification.",
"recommended_action": "Terminate the call and verify the claim through an official number you find independently.",
"evidence": {
"belief_installation": "False claim of tax discrepancies and an arrest warrant.",
"verification_suppression": "Instructs the target not to contact anyone else.",
"urgency_fabrication": "Immediate payment demanded.",
"authority_hijacking": "Claims to be from the IRS Criminal Investigation Division.",
"emotional_flooding": "Threat of arrest creates fear.",
"exit_path_closure": "Explicit instruction not to hang up."
}
}
5.3 GBThis model is intended to help users recognize manipulative scam behavior. It should be deployed with:
If you use this model, cite the project repository and hackathon submission:
@misc{aegis_gemma4_2026,
title={AEGIS: Adaptive Epistemic Guard for Intelligent Scam Defense},
author={GOVINDFROM},
year={2026},
howpublished={Hugging Face model repository}
}