destin-corporate-travel

Destin - Corporate Travel AI Assistant

<div align="center"> <img src="https://img.shields.io/badge/Model-Phi--3.5--mini-blue" alt="Model"> <img src="https://img.shields.io/badge/Task-Travel%20Assistant-green" alt="Task"> <img src="https://img.shields.io/badge/Language-English-red" alt="Language"> </div>

🏨 Model Description

Destin is an AI-powered corporate travel assistant specialized in hotel search and booking. Built on Microsoft's Phi-3.5-mini-instruct and fine-tuned with LoRA, Destin understands natural language hotel queries and converts them into structured booking data.

Key Features

• 🎯 Natural Language Understanding: Parse complex hotel queries in conversational language
• 🏢 Corporate Travel Focus: Optimized for business travel scenarios
• 📊 Structured Output: Generates clean JSON for booking systems
• 💬 Conversational: Handles greetings, follow-ups, and clarifications
• ⚡ Fast: Optimized for production deployment

🚀 Quick Start

Installation

pip install transformers torch accelerate

Basic Usage

from transformers import AutoTokenizer, AutoModelForCausalLM
import torch

# Load model
model_name = "nathishdev/destin-corporate-travel"
tokenizer = AutoTokenizer.from_pretrained(model_name, trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained(
    model_name,
    torch_dtype=torch.float16,
    device_map="auto",
    trust_remote_code=True
)

# Generate response
def chat(query):
    prompt = f'''<|system|>
You are Destin, a corporate travel assistant.<|end|>
<|user|>
{query}<|end|>
<|assistant|>
'''
    
    inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
    outputs = model.generate(
        **inputs,
        max_new_tokens=300,
        temperature=0.7,
        top_p=0.9,
        do_sample=True
    )
    
    response = tokenizer.decode(outputs[0], skip_special_tokens=True)
    # Extract assistant response
    response = response.split("<|assistant|>")[-1].strip()
    
    return response

# Example queries
print(chat("Hi! Who are you?"))
# Output: I'm Destin, an AI assistant specialized in corporate hotel bookings...

print(chat("hotel bangalore whitefield under 6k dec 10-12"))
# Output: {"destination": {"city": "Bangalore", "area": "Whitefield"}, ...}

💡 Use Cases

1. Hotel Search Parsing

query = "urgent booking mumbai bkc tonight under 8000"
result = chat(query)
# Extracts: city, area, date, budget, urgency

2. Conversational Booking

query = "I need a hotel in Delhi"
response = chat(query)
# Asks for: dates, budget, area preferences

3. Multi-City Travel

query = "hotels in bangalore dec 10-12 and mumbai dec 13-15, budget 6k each"
result = chat(query)
# Parses both destinations

📊 Model Details

Architecture

• Base Model: microsoft/Phi-3.5-mini-instruct (3.8B parameters)
• Fine-tuning Method: LoRA (Low-Rank Adaptation)
• Trainable Parameters: 8.9M

Performance

• Inference Speed: 30-40 tokens/second (FP16 on A10G GPU)
• Memory Usage: ~8GB GPU RAM (FP16)
• Response Time: 5-10 seconds per query

Training Dataset

The model was trained on:

• Identity & personality examples (30)
• Hotel query parsing (80)
• Conversational flows (20)
• Edge cases (typos, vague queries) (20)

🎯 Capabilities

What Destin Can Do

✅ Parse hotel search queries into structured JSON
✅ Handle natural language (typos, abbreviations)
✅ Understand dates (relative and absolute)
✅ Extract location, budget, guests, urgency
✅ Conversational follow-ups
✅ Multi-city trip planning

Supported Formats

Input Examples:

• "hotel bangalore under 5k next week"
• "urgent booking mumbai tonight"
• "need accommodation in delhi aerocity for 3 nights"
• "cheap hotel gurgaon cyber city for 2 people"

Output Format:

{
  "destination": {"city": "Bangalore", "area": "Whitefield"},
  "dates": {"check_in": "2024-12-10", "check_out": "2024-12-12"},
  "budget": {"max_per_night": 6000, "currency": "INR"},
  "guests": {"adults": 1}
}

🔧 Integration Examples

FastAPI Deployment

from fastapi import FastAPI
from pydantic import BaseModel

app = FastAPI()

class QueryRequest(BaseModel):
    query: str

@app.post("/parse")
async def parse_query(request: QueryRequest):
    response = chat(request.query)
    return {"response": response}

MCP (Model Context Protocol)

# Compatible with MCP for agent workflows
# Use Destin as a tool in multi-agent systems

📈 Roadmap

• Flight booking support
• Cab/transport integration
• Multi-language support
• International destinations
• Real-time price integration

🤝 Contributing

Feedback and suggestions welcome! Open an issue or PR.

📄 License

MIT License - Free for commercial use

Acknowledgments

• Base model: Microsoft Phi-3.5-mini-instruct
• Training platform: Kaggle (free GPU)
• Fine-tuning: LoRA + BitsAndBytes

---

Built with ❤️ for corporate travelers