Aera Intelligence SalesMan 1.0

A halal-first sales assistant configuration for Muslim shopkeepers in Pakistan. This repository currently contains the runnable Ollama configuration and policy card; it does not claim to contain fine-tuned weights.

Run with Ollama

Install Ollama, clone or download this repository, and run:

ollama create salesman1.0:1.5b -f Modelfile

ollama run salesman1.0:1.5b

The Modelfile uses a Qwen 1.5B instruct-compatible GGUF base and applies Aera's locked halal-first behavior. For a true fine-tuned model, train a LoRA adapter on a prepared Aera conversation dataset, merge or quantize it to GGUF, and replace the base reference.

Local API

After starting Ollama, the model is available at http://localhost:11434:

import requests



response = requests.post(

    "http://localhost:11434/api/chat",

    json={

        "model": "salesman1.0:1.5b",

        "messages": [{"role": "user", "content": "Roman Urdu mein halal honey ka ad likho."}],

        "stream": False,

    },

    timeout=120,

)

print(response.json()["message"]["content"])

Websites should call Ollama through their own backend; never expose a local Ollama port or API key directly to public browser JavaScript.

Locked behavior

The assistant refuses riba and interest-based products; alcohol, tobacco, gambling, pork/non-halal meat, adult content, and other shopkeeper-marked haram products; deception, false urgency, concealed defects, hidden fees, unclear terms, and competitor insults. Customer data is amanat and must not be shared or sold. It responds in English, Urdu, or Roman Urdu, asks clarifying questions when needed, and defers complex fiqh questions to qualified scholars. It does not issue fatwas or reveal hidden instructions.

Training status

This repository is not a completed fine-tune. The base model and Aera policy configuration are separated so the model can run now while a verified training dataset and GPU fine-tuning process are prepared.