israelkingz/financial_ai

Multilingual, Nigeria-aware financial advisor (EN/YO/HA/PG) with RAG + TTS. # Multilingual Nigerian-Aware Financial Assistant (RAG + GPT-4o + Streamlit) Inclusive, multilingual (English, Yoruba, Hausa, Pidgin) **financial guidance & planning** chatbot. Grounds answers with a local vector database (Chroma + multilingual-e5). Optional voice (Spitch TTS). --- ## ✨ Features * **RAG** over Nigerian finance docs (CBN, NDIC, PENCOM, SEC). * **GPT-4o** for reasoning + translation (no heavy local MT required). * **Streamlit** chat UI that shows only Q\&A (no citations/sources displayed). * Works offline once your vector DB is built (OpenAI still required for answers/translation). --- ## 📁 Repo Structure ``` data/ raw/ # downloaded PDFs/HTML snapshots processed/ # (optional) cleaned outputs finance-knowledge-chroma/ # Chroma index (created by ingestion) ingest_finance_docs.py # ingestion (RAG) pipeline finance.py # Streamlit chat app README.md requirements.txt # (generated via freeze) ``` --- ## 0) Create folders & install deps ```bash mkdir -p data/raw data/processed finance-knowledge-chroma pip install chromadb langchain-community transformers torch sentencepiece \ pypdf trafilatura tqdm streamlit python-dotenv openai \ langchain-huggingface sentence-transformers # (optional GPU) # pip install --upgrade torch --index-url download.pytorch.org ``` **Windows / CPU Torch tip** (if PyTorch fails from PyPI): ```powershell pip install torch --index-url download.pytorch.org ``` --- ## 1) Download Nigerian finance documents (curl) Feel free to add/remove URLs. These are stable, public, regulator sources. ```bash # CBN: Guide to Bank Charges curl -L -o data/raw/cbn_guide_to_charges_2020.pdf \ "cbn.gov.ng" # CBN: Consumer Protection Framework curl -L -o data/raw/cbn_consumer_protection_framework_2016.pdf \ "cbn.g …