📄 Quotation Generation Agent with AI Tool-Calling Architecture

An intelligent, autonomous Quotation Generation Agent built with Flask, Local Hugging Face LLMs, SQLite, and ReportLab.

The system demonstrates a robust AI Tool-Calling Architecture: the language model does not directly calculate prices or hallucinate inventory. Instead, it acts as an orchestrator that selects appropriate Python tools, executes business actions against the SQLite catalog and user session, and renders professional, downloadable PDF quotations.

---

🌟 Architecture & Core Workflow

User ──► Flask UI / API ──► QuotationAgent ──► Local Hugging Face LLM
                                   │
                                   ▼
                              ToolRegistry
                                   │
        ┌──────────────────────────┼──────────────────────────┐
        ▼                          ▼                          ▼
  search_products            add_to_cart                generate_pdf
   (SQLite DB)              (Session Cart)           (ReportLab Engine)

Execution Flow:

1. User Request: The user interacts conversationally (e.g., "Search for laptops under $1000 and add the first one to my cart").
2. Orchestration: QuotationAgent passes the conversation history and OpenAI-compatible tool definitions to the LLM (Qwen/Qwen3-1.7B or offline fallback).
3. Tool Invocation: The LLM responds with structured tool calls (e.g., search_products(category='laptop', budget=1000)).
4. Deterministic Execution: ToolRegistry invokes the Python function with the active ToolContext (connecting the session, SQLite database, and PDF generator).
5. Observation Loop: The deterministic result is fed back into the agent history (role: "tool").
6. Final Synthesis: The agent produces a natural-language confirmation and quotation download link.

---

🚀 Key Features

• AI Tool-Calling Agent Loop: Autonomous multi-step ReAct loop (up to 8 steps) with JSON normalization and error recovery.
• Zero Financial Hallucinations: Product prices, inventory levels, GST/tax rates, and shipping calculations are 100% deterministic, governed by Python and SQLite.
• On-Premise / Local Inference: Runs with Hugging Face Transformers (Qwen/Qwen3-1.7B) with graceful fallback mode when offline.
• Dynamic PDF Generation: Automatically outputs structured B2B/B2C PDF quotations using ReportLab with unique IDs (e.g., QT-20260909-XXXXXX).
• Session-Based Cart Management: Stateful cart persistence across multi-turn interactions.
• Dual Interface: Web interface (/, /search, /ai-search, /cart, /quotation) plus JSON REST API (/api/agent).

---

📁 Project Structure

chatbot/
├── app.py                 # Core Flask application, web routes & REST API (/api/agent)
├── agent.py               # Autonomous QuotationAgent loop and response normalization
├── tool_registry.py       # ToolRegistry and ToolSpec dataclass (OpenAI schema converter)
├── tools.py               # Business tools (search_products, add_to_cart, generate_pdf, etc.)
├── llm.py                 # Hugging Face local LLM adapter & smart heuristic fallback
├── chatbot.py             # Deterministic NLP parser and conversation memory summarizer
├── database.py            # SQLite DatabaseManager with seed product catalog
├── models.py              # Product and CartItem data models
├── pdf_generator.py       # ReportLab PDF compilation engine
├── products.db            # SQLite database file
├── requirements.txt       # Project dependencies
├── templates/             # HTML Templates (index.html, cart.html, quotation.html)
├── pdfs/                  # Generated PDF quotation documents
└── tests/                 # Automated unit and integration test suite
    ├── test_agent_loop.py
    ├── test_conversation_memory.py
    └── test_tools_and_api.py

---

🛠️ Tool Registry Reference

All tools are registered in tool_registry.py and exposed via standard OpenAI JSON schemas:

Tool Name	Parameters	Description
search_products	category (str), filters (dict), budget (float)	Searches the catalog for matching items within budget
add_to_cart	product_id (int), quantity (int)	Adds or consolidates product quantity in the session cart
remove_from_cart	product_id (int)	Removes specified item from the shopping cart
update_quantity	product_id (int), quantity (int)	Modifies quantity; removes item if quantity is set to 0
show_cart	None	Returns items, subtotal, GST (tax), and shipping estimates
generate_quotation	None	Formulates complete quotation breakdown
generate_pdf	None	Generates a styled ReportLab PDF quotation in /pdfs/
clear_cart	None	Clears all items from the current session cart

---

⚡ Quickstart & Installation

1. Prerequisites

• Python 3.9, 3.10, or 3.11 installed.

2. Setup Virtual Environment

# Clone the repository
git clone https://github.com/AgentRoomAI/Quotation-Generation-Agent.git
cd Quotation-Generation-Agent/chatbot

# Create virtual environment
py -3.11 -m venv .venv

# Activate virtual environment
# Windows:
.venv\Scripts\activate
# Linux/macOS:
source .venv/bin/activate

# Install dependencies
pip install -r requirements.txt

3. Run Automated Tests

Verify that all 14 unit and integration tests pass:

pytest tests -v

4. Launch the Web Application

python app.py

Open your browser at http://127.0.0.1:5000.

---

🔌 REST API Usage

You can interact with the agent programmatically via POST /api/agent:

Request

curl -X POST http://127.0.0.1:5000/api/agent \
  -H "Content-Type: application/json" \
  -d '{"message": "search laptop"}'

Response

{
  "status": "success",
  "reply": "I found the following products: #18 Acer Aspire 7 ($75999.0); #20 Apple MacBook Air M2 ($119999.0)...",
  "cart": [],
  "summary": {"subtotal": 0.0, "gst_total": 0.0, "shipping": 0.0, "grand_total": 0.0},
  "quotation_id": null,
  "quotation_pdf": null
}

---

📜 License

Developed for educational, research, and business automation purposes.