CHATMOBOT/mobot-ghana-complete

MOBOT - Ghana Financial Assistant 🤖🇬🇭

Your friendly financial assistant for Ghana

---

🎯 What is MOBOT?

MOBOT is a multilingual, multimodal conversational AI specialized for Ghanaian financial services. Built on Mistral-7B, MOBOT helps users with:

• 💰 Money Transfers - Send money to friends and family
• 📄 Bill Payments - Pay ECG, GWCL, hospital bills, and more
• 📊 Financial Tracking - Check balances, view spending summaries
• 💡 Financial Advice - Savings tips, debt planning, budgeting
• 🎤 Voice Interaction - Speak naturally in Twi, Pidgin, English, and more
• 📷 Bill Scanning - Upload bill photos for automatic processing

---

✨ Key Features

🌍 Multilingual Support

• English - Full support
• Twi - Native language support
• Pidgin - Ghanaian Pidgin English
• Hausa - Northern Ghana language
• Ga - Greater Accra region
• Ewe - Volta region

🎨 Multimodal Capabilities

• 📝 Text - Natural language conversations
• 🎤 Voice Input - Speech-to-text (STT) with Whisper
• 📷 Image Processing - OCR for bills and receipts
• 🔊 Voice Reply - Text-to-speech (TTS) responses

💼 Financial Services

• Send money / transfers
• Pay utility bills (ECG, GWCL)
• Buy airtime and data
• Check account balance
• View transaction history
• Get spending summaries
• Financial coaching

🤝 Empathetic & Safe

• Emotional intelligence for sensitive situations
• Safety patterns for sensitive information
• Consent management for transactions
• Audit logging for compliance

---

🚀 Quick Start

Installation

pip install transformers torch

Basic Usage

from transformers import AutoModelForCausalLM, AutoTokenizer

# Load model
model_name = "CHATMOBOT/mobot-ghana-complete"
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(model_name)

# Chat with MOBOT
user_input = "Hey Mobot, send 50 cedis to Ama"
prompt = f"<s>[INST] You are MOBOT, Ghana's friendly financial assistant.\n\nUser: {user_input} [/INST]"

inputs = tokenizer(prompt, return_tensors="pt")
outputs = model.generate(**inputs, max_new_tokens=512, temperature=0.2)
response = tokenizer.decode(outputs[0], skip_special_tokens=True)

print(response)

With Pipeline

from transformers import pipeline

mobot = pipeline(
    "text-generation",
    model="CHATMOBOT/mobot-ghana-complete"
)

response = mobot(
    "User: How much did I spend this week?",
    max_new_tokens=512,
    temperature=0.2
)

---

📚 Examples

Money Transfer

User: Send 50 cedis to Kwame
Mobot: Sure! I'll send ₵50 to Kwame. Please confirm?

Bill Payment

User: [Uploads ECG bill photo]
Mobot: I found an ECG bill for ₵100.00 (ref: REF123). Would you like me to pay it?

Voice Note (Pidgin)

User: [Voice] "Chale send 20 ghs to Ama"
Mobot: Okay boss, sending ₵20 to Ama. Please confirm?

Financial Advice

User: Help me save money
Mobot: Great question! Here are some tips: 
1. Set a monthly savings goal
2. Automate transfers to savings
3. Track your spending
Want more advice?

---

🔌 Integrations

WhatsApp

# See: integrations/whatsapp_webhook_example.py
# Full WhatsApp Business API integration

Telegram

# See: integrations/telegram_bot_example.py
# Interactive bot with buttons

Agent Orchestration

# See: agents/orchestrator.py
# Multi-step workflows

---

📊 Model Details

Architecture

• Base Model: Mistral-7B-v0.1
• Fine-tuning: LoRA (Low-Rank Adaptation)
• LoRA Rank: 128
• LoRA Alpha: 32
• Parameters: 7B
• Context Length: 2048 tokens

Training

• Current Dataset: 15,264 examples (quality-filtered)
• Total Available: 509,438+ examples (from larger dataset)
• Languages: 6 languages (English, Twi, Pidgin, Hausa, Ga, Ewe)
• Intents: 12+ financial intents
• Training Time: ~48 hours on RTX 4090
• Training Config: See training_config.yaml

Performance

• Intent Accuracy: ≥95%
• Slot F1: ≥92%
• Empathy Score: ≥4.0/5.0

---

🛠️ Advanced Usage

Multimodal Processing

# OCR for bills
from scripts.multimodal.ocr_receipts import process_image
parsed = process_image("bill.jpg")

# STT for voice
from scripts.multimodal.stt_whisper import transcribe_audio
transcript = transcribe_audio("voice_note.mp3")

# TTS for replies
from services.tts_service import synthesize_text
audio = synthesize_text("Hello! I'm MOBOT.")

Agent Orchestration

from agents.orchestrator import AgentOrchestrator

orchestrator = AgentOrchestrator()
plan = orchestrator.plan("Pay this bill", context={"image": "bill.jpg"})
result = orchestrator.execute(plan, user_id="user123")

---

📖 Documentation

• Model Card - Detailed model information
• Phase 4 Implementation - Multimodal features
• Complete System Status - Full system overview
• Training Guide - Training instructions

---

🔒 Safety & Privacy

• ✅ Refusal patterns for sensitive information
• ✅ PII redaction in training data
• ✅ Consent management for transactions
• ✅ Audit logging for compliance
• ✅ Ghana Data Protection Act compliant

---

🤝 Contributing

We welcome contributions! See our GitHub repository for:

• Issue reporting
• Feature requests
• Pull requests
• Documentation improvements

---

📄 License

Apache 2.0 License

---

🙏 Acknowledgments

• Mistral AI - Base model
• RunPod - Training infrastructure
• Coqui TTS - Voice synthesis
• OpenAI Whisper - Speech recognition
• PaddleOCR - Image processing

---

📞 Contact

• Email: [email protected]
• Info: [email protected]
• GitHub Issues: https://github.com/your-org/mobot/issues

🌐 Social Media

• Facebook: MOBOT AI Ghana
• Instagram: @MOBOTAI_GHANA
• X (Twitter): @MOBOTAI_GHANA
• WhatsApp Channel: MOBOT Channel

---

Made with ❤️ for Ghana

🇬🇭 MOBOT - Your Financial Assistant