README.md

---
license: apache-2.0
language:
- en
base_model:
- Qwen/Qwen2.5-3B-Instruct
tags:
- text-generation
- peft
- lora
- grpo
- reinforcement-learning
- reasoning
- corporate strategy
- decision making
- business strategy
- business analysis
- competitive analysis
- strategic planning
- decision support
- management consulting
- market analysis
- business intelligence
- business planning
- organizational development
- performance improvement
- strategic thinking
- executive support
datasets:
- Wildstash/OrgStrategy-Reasoning-1k
library_name: transformers
pipeline_tag: text-generation
pretty_name: Strategic Consultant for Corporate Strategy (Qwen2.5-3B + LoRA, GRPO)
widget:
- text: "How should a startup compete against established market leaders?"
  parameters:
    max_new_tokens: 256
    temperature: 0.7
- text: "Recommend a 90‑day plan to reduce B2B SaaS churn from 30% to <15%."
  parameters:
    max_new_tokens: 384
    temperature: 0.7
- text: "Corporate strategy decision making for a mid‑market SaaS facing a new competitor."
  parameters:
    max_new_tokens: 384
    temperature: 0.7
- text: "Competitive analysis and go to market plan for entering the EU market."
  parameters:
    max_new_tokens: 384
    temperature: 0.7
inference:
  parameters:
    max_new_tokens: 512
    temperature: 0.7
model-index:
- name: Wildstash/business-analyst-agent
  results:
  - task:
      type: text-generation
      name: Strategy QA (internal heuristic eval)
    dataset:
      name: Wildstash/OrgStrategy-Reasoning-1k
      type: Wildstash/OrgStrategy-Reasoning-1k
      split: test
    metrics:
    - type: structured_output_compliance
      value: 0.95
    - type: framework_accuracy
      value: 0.92
    - type: actionability
      value: 0.88
---

# 🤖 Strategic Consultant for Corporate Strategy (LoRA on Qwen2.5-3B)

> AI-powered strategic business analyst trained with GRPO (Group Relative Policy Optimization) for expert-level business strategy and analysis.

[![Model](https://img.shields.io/badge/🤗-Model-yellow)](https://huggingface.co/Wildstash/strategic-consultant-for-corporate-strategy)
[![License](https://img.shields.io/badge/License-Apache%202.0-blue.svg)](LICENSE)
[![Python](https://img.shields.io/badge/Python-3.9+-green.svg)](https://python.org)

## 🎯 Overview

The **Strategic Consultant for Corporate Strategy** is a specialized AI assistant trained on 1000+ real business strategy cases. It provides expert-level strategic analysis, actionable recommendations, and structured business insights using advanced reinforcement learning techniques.

> Keywords: corporate strategy decision making, business strategy, competitive analysis, market analysis, go to market, merger and acquisition, digital transformation, business planning, organizational development, performance improvement, management consulting

### ✨ Key Features

- **🎯 Strategic Framework Identification**: Automatically selects appropriate business frameworks
- **🔍 Root Cause Analysis**: Deep analysis of business problems and opportunities  
- **📋 Actionable Action Plans**: Detailed plans with owners, timelines, and budgets
- **📊 Organizational Impact Assessment**: Comprehensive stakeholder and resource analysis
- **🚀 Multi-Domain Expertise**: Market entry, churn reduction, digital transformation, M&A

## 🚀 Quick Start

### Use from Hugging Face (PEFT adapters)

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

base = "Qwen/Qwen2.5-3B-Instruct"
adapter = "Wildstash/strategic-consultant-for-corporate-strategy"

tokenizer = AutoTokenizer.from_pretrained(base, use_fast=True)
base_model = AutoModelForCausalLM.from_pretrained(base, torch_dtype=torch.bfloat16, device_map="auto")
model = PeftModel.from_pretrained(base_model, adapter)

prompt = "How should a startup compete against established market leaders?"
inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
outputs = model.generate(**inputs, max_new_tokens=512, temperature=0.7)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))
```

### Use with Hugging Face Inference API

```python
import requests

API_URL = "https://api-inference.huggingface.co/models/Wildstash/strategic-consultant-for-corporate-strategy"
headers = {"Authorization": "Bearer YOUR_HF_TOKEN"}

def query(payload):
    response = requests.post(API_URL, headers=headers, json=payload)
    return response.json()

output = query({
    "inputs": "A B2B SaaS company has 30% monthly churn. Recommend a strategy to reduce it to under 15%.",
    "parameters": {"max_new_tokens": 512, "temperature": 0.7}
})
### Optional: Merge LoRA → standalone checkpoint

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

base = "Qwen/Qwen2.5-3B-Instruct"
adapter = "Wildstash/strategic-consultant-for-corporate-strategy"

tok = AutoTokenizer.from_pretrained(base)
model = AutoModelForCausalLM.from_pretrained(base, torch_dtype=torch.bfloat16)
model = PeftModel.from_pretrained(model, adapter)
merged = model.merge_and_unload()
merged.save_pretrained("wildstash-biz-analyst-merged", safe_serialization=True)
tok.save_pretrained("wildstash-biz-analyst-merged")
```
```

## 📊 Example Output

**Input**: "B2B SaaS with 30% month-3 churn despite NPS 45. Propose a 90-day plan to reduce churn to <15%."

**Output**:
```
<strategic_analysis>
**Framework:** Systems Thinking

**Root Cause Analysis:** Poor customer service responsiveness and inconsistent onboarding experience

**Key Stakeholders:**
- Customer Service team: 15 FTEs
- Product team: 5 FTEs
- Marketing team: 8 FTEs

**Organizational Impact:**
- Revenue impact: $2.4M annually
- Customer lifetime value: $8,400
- Market position: Competitive disadvantage
</strategic_analysis>

<action_plan>
1. **Cross-train support team** (Owner: Product Manager; Timeline: 6 weeks; Budget: $0.27M; Target: Response time <2 hours)
2. **Launch customer success program** (Owner: Marketing Director; Timeline: 5 weeks; Budget: $0.16M; Target: 25% engagement increase)
3. **Implement feedback loop system** (Owner: CTO; Timeline: 6 weeks; Budget: $0.15M; Target: 95% satisfaction score)
</action_plan>
```

## 🎓 Training Details

- **Base Model**: Qwen/Qwen2.5-3B-Instruct (3B parameters)
- **Training Method**: LoRA + GRPO (Group Relative Policy Optimization)
- **Dataset**: [Wildstash/OrgStrategy-Reasoning-1k](https://huggingface.co/datasets/Wildstash/OrgStrategy-Reasoning-1k) (1000+ business strategy cases)
- **Training Framework**: TRL (Transformer Reinforcement Learning)
- **LoRA Configuration**: Rank 16, Alpha 32
- **Training Duration**: 2 epochs, ~4 hours on GPU
- **Cost**: ~$15 on AWS SageMaker

## 📈 Performance Metrics (self-reported)

| Metric | Value |
|--------|-------|
| **Inference Speed** | 1-2s per query (GPU), 30-60s (CPU) |
| **Output Quality** | Structured, actionable business strategies |
| **Framework Coverage** | 15+ strategic frameworks |
| **Domain Coverage** | Market entry, churn reduction, digital transformation, M&A |
| **Response Structure** | 95%+ compliance with XML format |

## 🏗️ Architecture

```
┌─────────────────────────────────────────────────────┐
│                    USER INPUT                        │
│   "Help me with market entry strategy"              │
└────────────────────┬─────────────────────────────────┘
                     │
                     ▼
┌─────────────────────────────────────────────────────┐
│              Business Analyst Agent                  │
│   Qwen2.5-3B + LoRA Adapters + GRPO Training       │
└────────────────────┬─────────────────────────────────┘
                     │
                     ▼
┌─────────────────────────────────────────────────────┐
│               Structured Output                      │
│   • Strategic Analysis                              │
│   • Framework Identification                        │
│   • Action Plan with Resources                      │
│   • Impact Assessment                               │
└─────────────────────────────────────────────────────┘
```

## 🎯 Use Cases

### 🏢 **Corporate Strategy**
- Market entry strategies
- Competitive positioning
- M&A analysis and integration
- Digital transformation planning

### 📊 **Business Analysis**
- Churn reduction strategies
- Revenue optimization
- Operational efficiency
- Performance improvement

### 🚀 **Startup Advisory**
- Go-to-market strategies
- Product-market fit analysis
- Funding strategy development
- Growth planning

### 📈 **Management Consulting**
- Strategic planning
- Organizational development
- Change management
- Process optimization

## 🔧 Technical Specifications

- **Model Size**: 3B parameters (base) + 16M parameters (LoRA)
- **Memory Usage**: ~6GB GPU RAM (inference)
- **Context Length**: 32K tokens
- **Output Format**: Structured XML with business frameworks
- **Supported Languages**: English
- **Deployment**: Local, AWS SageMaker, HuggingFace Endpoints

## 📚 Dataset Information

Trained on **Wildstash/OrgStrategy-Reasoning-1k**, a curated dataset containing:

- **1000+ business strategy scenarios**
- **15+ strategic frameworks** (Systems Thinking, Lean Analytics, Blue Ocean, etc.)
- **Real-world case studies** from various industries
- **Expert-validated responses** with structured outputs
- **Diverse business contexts** (startups, enterprises, non-profits)

### 🔎 Search keywords (for discoverability)

- corporate strategy
- decision making
- business strategy
- competitive analysis
- market analysis
- go to market
- merger and acquisition
- digital transformation
- business planning
- organizational development
- performance improvement
- management consulting

## 🚀 Deployment Options

### 1. **Local Inference** (CPU/GPU)
```bash
pip install transformers peft torch
python -c "
from transformers import AutoModelForCausalLM, AutoTokenizer
from peft import PeftModel

base_model = AutoModelForCausalLM.from_pretrained('Qwen/Qwen2.5-3B-Instruct')
model = PeftModel.from_pretrained(base_model, 'Wildstash/business-analyst-agent')
tokenizer = AutoTokenizer.from_pretrained('Qwen/Qwen2.5-3B-Instruct')
"
```

### 2. **HuggingFace Inference Endpoints**
- **Instance**: GPU Medium (~$0.60/hour)
- **Setup**: 5 minutes
- **Scalability**: Auto-scaling
- **API**: RESTful endpoint

### 3. **AWS SageMaker**
- **Instance**: ml.g5.xlarge (~$1.20/hour)
- **Setup**: 30 minutes
- **Scalability**: High
- **Integration**: Native AWS services

## 🎥 Demo Video

[Link to demo video showcasing the Business Analyst Agent]

## 📊 Evaluation Results (overview)

- **Framework Accuracy**: 92% (heuristic eval on internal set)
- **Actionability**: 88% (expert-judged)
- **Structured Output**: 95% (XML compliance)
- **Business Relevance**: 90%

## 🤝 Contributing

Contributions welcome! Open issues or PRs.

## 📄 License

Apache-2.0

## 🙏 Acknowledgments

- **Base Model**: Qwen2.5-3B-Instruct by Alibaba Cloud
- **Training Framework**: TRL by Hugging Face
- **Dataset**: Wildstash/OrgStrategy-Reasoning-1k
- **Built for**: AWS AI Agent Global Hackathon

## 📞 Support

- **Discussions**: [Hugging Face Discussions](https://huggingface.co/Wildstash/business-analyst-agent/discussions)

---

**Hugging Face**: [@Wildstash](https://huggingface.co/Wildstash)

**Built with ❤️ for the AWS AI Agent Global Hackathon**