title	KMITL Budget AI Chatbot
emoji	🎓
colorFrom	indigo
colorTo	blue
sdk	docker
app_file	streamlit_app.py
app_port	8501
pinned	false

🎓 KMITL Budget AI Chatbot (CEIPP)

"Intelligent Compliance & Disbursement Analysis for Institutional Finance"

CEIPP is an advanced AI assistant designed to bridge the gap between complex government procurement regulations and institutional operational efficiency at King Mongkut's Institute of Technology Ladkrabang (KMITL). By leveraging Generative AI and Retrieval-Augmented Generation (RAG), it automates receipt verification and provides instant, source-grounded regulatory consultation.

---

🌟 Core Intelligence Modules

1. 👁️ Smart OCR Auditor (Vision AI)

• Multimodal Extraction: Utilizes Google Gemini 1.5 Flash to extract high-fidelity Thai text from receipts and tax invoices.
• 5-Point Compliance Check: Automated validation of Vendor details, Transaction dates, Total amounts, Buyer identification, and mandatory Signatures/Stamps.
• Smart Priority Logic: Intelligent bypass for certified corporate vendors (e.g., Makro, 7-11), reducing administrative bottlenecks for computer-generated invoices.

2. 💬 Regulatory Consultant (RAG Chatbot)

• Extensive Knowledge Base: Indexed repository of 29 official KMITL and Ministry of Finance PDFs.
• Evidence-Grounded Answers: Every response includes Filename and Page Number citations to ensure maximum transparency.
• Thai-Optimized Retrieval: Custom regex-based text cleaning minimizes character corruption, ensuring precise semantic search results for the Thai language.

3. 📊 Admin Monitoring & AI Evaluation

• LLM-as-a-Judge Framework: Employs Gemma 3 4B to quantitatively evaluate AI performance across Faithfulness (4.51/5.0), Answer Relevance (4.45/5.0), and Context Precision (4.02/5.0).
• Benchmarked Performance: Achieving 91.64% Avg Recall@10 and 0.75 MRR across a comprehensive 323-item golden dataset.
• Real-time Analytics: Interactive dashboard visualizing user feedback trends, approval distributions, and RAG accuracy benchmarks.

---

🏗️ System Architecture

The project implements a hybrid offline/online pipeline to ensure both data privacy and high-speed inference.

graph TD
    User((User Interaction)) --> SL[Streamlit Mobile-First UI]
    SL --> Chat[Chatbot Interface]
    SL --> OCR[Receipt Verification]
    
    subgraph "AI Intelligence Layer"
        OCR -->|Vision| Gemini[Gemini 1.5/2.0 Flash]
        Chat -->|Prompt| OpenRouter[Gemma 3 27B-IT]
        OpenRouter --> Retriever[Semantic Retriever]
    end
    
    subgraph "Knowledge Retrieval (RAG)"
        Retriever --> ChromaDB[(ChromaDB Vector Store)]
        ChromaDB --> Docs[29 PDF Regulations]
    end
    
    subgraph "Automated Evaluation Model"
        Chat --> Judge[Gemma 3 4B-IT Judge]
        Judge --> Dashboard[Admin Monitoring Panel]
    end

Loading

---

🛠️ Technical Stack

Layer	Component	Technology
Frontend	UI Framework	Streamlit
Logic	Language Model	Gemma 3 (via OpenRouter API)
Vision	OCR Engine	Google Gemini 1.5 Flash (Native SDK)
Storage	Vector Database	ChromaDB (Persistent SQLite)
Language	Thai NLP	sentence-transformers/paraphrase-multilingual
Deployment	Infrastructure	Python 3.10+ / Streamlit Community Cloud

---

🚀 Installation & Setup

1. Prerequisite Environments

git clone https://github.com/Ratthabhumi/CEIPP.git
cd CEIPP
pip install -r requirements.txt

2. Configure API Credentials

Create a .streamlit/secrets.toml file or provide keys via the application sidebar:

OPENROUTER_API_KEY = "your_openrouter_key"
GEMINI_API_KEY = "your_gemini_key"

3. Initialize & Run

streamlit run app.py

Note: The system will automatically build the ChromaDB vector database upon first execution (requires 29 PDF documents in the ./Docs folder).

---

📑 Included Regulations (29 Documents)

The system currently covers a comprehensive set of KMITL and Public Procurement regulations, including:

• ✅ KMITL Procurement and Supplies Regulations (2560-2562)
• ✅ Research Grant Disbursement Rules
• ✅ Training & Seminar Expenditure Policies
• ✅ Official Travel Allowance Regulations
• ✅ Public Procurement and Supplies Management Act (2560)

---

📬 Maintenance & Contact

• Project Lead: Ratthabhumi/CEIPP Team
• Academic Context: 01276390 Computer Engineering Project Preparation (School of Engineering)
• Status: Final Production (Ready for Deployment)

---

Developed with ❤️ to modernize institutional finance through Artificial Intelligence.