---
base_model: meta-llama/Llama-3.2-3B-Instruct
library_name: peft
tags:
- hr-analytics
- attrition-risk
- reasoning-traces
- autoscientist
- llama-3.2
- peft
license: apache-2.0
datasets:
- CorpIntel-Attrition-Reasoning-v1
language:
- en
pipeline_tag: text-generation
---

# CorpIntel-HR-Agent

`CorpIntel-HR-Agent` is an instruction-tuned 3B parameter model fine-tuned on `meta-llama/Llama-3.2-3B-Instruct` using PEFT (LoRA). The model is specifically engineered to evaluate complex, multi-variable employee telemetry (commute friction, salary hikes, overtime, job satisfaction, and career stagnation) to perform attrition risk modeling and generate structured **Managerial Intervention Plans**.

Unlike standard binary classifiers that output a simple "Yes/No" risk score, `CorpIntel-HR-Agent` generates explicit step-by-step reasoning traces detailing *why* an employee is a flight risk and *what* specific managerial steps can retain them.

## Model Details

### Model Description

- **Developed by:** Asad Ullah Dogar
- **Model type:** PEFT / LoRA Adapter (Causal LM)
- **Language(s):** English (en)
- **License:** Apache-2.0
- **Finetuned from model:** `meta-llama/Llama-3.2-3B-Instruct`
- **Platform:** Adaption Labs AutoScientist Engine

### Model Sources

- **Dataset:** `CorpIntel-Attrition-Reasoning-v1`
- **Base Architecture:** Llama 3.2 3B Instruct

---

## Uses

### Direct Use
* **HR Analytics & Decision Support:** Evaluating employee telemetry to identify hidden burnout and retention risks.
* **Reasoning Generation:** Synthesizing multi-variable data (e.g., long commute + low salary hike + high overtime) into actionable narrative diagnostics.
* **Retention Strategy Generation:** Crafting tailored Managerial Intervention Plans for HR Business Partners and regional leaders.

### Out-of-Scope Use
* **Automated Termination/Hiring:** This model is designed purely as an analytical decision-support tool. It must **not** be used for fully automated HR decisions without human oversight.

---

## Bias, Risks, and Limitations

* **Domain Specificity:** Trained on corporate HR telemetry schemas. Metrics using significantly different column formats may require input rephrasing or context mapping.
* **Decision Support Only:** The model outputs recommendations and reasoning traces based on input parameters; human HR expertise is required to validate intervention strategies.

---

## How to Get Started with the Model

```python
import torch
from transformers import AutoModelForCausalLM, AutoTokenizer
from peft import PeftModel

base_model_id = "meta-llama/Llama-3.2-3B-Instruct"
adapter_id = "CorpIntel-HR-Agent"  # Replace with your Hugging Face username/repo

tokenizer = AutoTokenizer.from_pretrained(base_model_id)
base_model = AutoModelForCausalLM.from_pretrained(
    base_model_id,
    torch_dtype=torch.bfloat16,
    device_map="auto"
)

model = PeftModel.from_pretrained(base_model, adapter_id)

prompt = """<|start_header_id|>system<|end_header_id|>
You are an elite HR Business Partner AI. Your objective is to evaluate heterogeneous employee telemetry to model attrition risk and generate a Managerial Intervention Plan.<|eot_id|>
<|start_header_id|>user<|end_header_id|>
Evaluate Candidate: Sales Executive | Travel: Frequently | Distance From Home: 24 miles | Monthly Income: 3200 | OverTime: Yes | JobSatisfaction: 1 | YearsSinceLastPromotion: 4<|eot_id|>
<|start_header_id|>assistant<|end_header_id|>"""

inputs = tokenizer(prompt, return_tensors="pt").to("cuda")
outputs = model.generate(**inputs, max_new_tokens=512, temperature=0.3)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))