Baraka is a modern, light-themed multilingual customer support assistant for Kenyan retail banking and SACCOs. It combines:

• FAQ retrieval (custom FAQs + a base banking dataset),
• automatic complaint routing to departments (ATM, Cards, Loans, Transfers, etc.),
• an Admin Console for managing FAQs and complaint tickets,
• multilingual chat with optional language locking (e.g. “reply in Kiswahili”).

Baraka is designed to avoid sensitive credential collection (no PINs, OTPs, passwords).

• Auto-detects language (when OpenAI is enabled)
• Replies in the user’s language
• You can force language in chat:
  reply in Kiswahili, reply in Somali, etc.

• Customers submit complaints via the Complaint page
• Baraka routes to a department automatically using:
  • keyword rules, then
  • TF-IDF similarity routing
• Saves each complaint into a local SQLite database (bankbot.db)

Baraka answers in this order:

1. Custom FAQs (admin-created, per department)
2. Base banking dataset (Bitext retail banking dataset from HF)
3. OpenAI fallback (only if enabled and only using top retrieved context)

Admins can:

• Add / edit / delete FAQs
• View and manage complaint tickets (status, priority, internal notes)
• View chat logs (source + similarity scores)

Includes a glassy, modern, light UI with BaseWeb styling overrides (selectboxes, menus, inputs).

• Streamlit UI
• SQLite persistence
• Pandas for data handling
• scikit-learn TF-IDF retrieval + cosine similarity
• Hugging Face dataset (parquet) as base FAQ dataset
• OpenAI API for translation + fallback answering

• app.py — the entire Streamlit app
• bankbot.db — auto-created SQLite database (users, FAQs, complaints, chat logs)

Install dependencies:

pip install streamlit openai pandas scikit-learn pyarrow huggingface-hub fsspec