[rish-anu]

dozer-1690-demo.mp4

/claim #1690
/claim getdozer/dozer#1690

Fixes getdozer/dozer#1690

Contributor & Payout Details
Contributor: Rish Anu

PayPal: srivikas346@gmail.com

Summary
Adds a complete, reproducible usecases/llm-banking-chatbot sample based on the Dozer LLM Chatbot hyper-personalization article:

Multi-Dataset Dozer Pipeline (dozer-config.yaml & dozer-config.postgres.yaml): Ingests customers.csv, credit_accounts.csv, transactions.csv, and card_products.csv via Dozer LocalStorage (and includes docker-compose.yml + sql/init_postgres.sql for PostgreSQL logical CDC).

Streaming SQL Unified Customer Profile: Joins customer demographics, credit account balances/utilization, and multi-category transaction aggregations (travel, dining, groceries, fuel, online_shopping, utilities) into /customer_profiles and /category_spending REST (:8080) and gRPC (:50051) endpoints.

Vector Database + LangChain Integration (app/vector_store.py, app/dozer_client.py, app/banking_chatbot.py): Performs semantic retrieval over credit card brochures, checks credit score and income eligibility against the live Dozer profile, and synthesizes hyper-personalized recommendations via LangChain (ChatOpenAI when OPENAI_API_KEY is set, plus deterministic --offline-fixture mode).

Real-Time CDC Transaction Shift (--simulate-realtime-tx): Demonstrates how streaming new international travel transactions into Dozer immediately updates the unified customer profile and dynamically flips the top recommended credit card.

Verification

python3 -m unittest discover -s usecases/llm-banking-chatbot/tests -v
python3 usecases/llm-banking-chatbot/app/banking_chatbot.py --offline-fixture --customer C001 --question "Which card is best for travel and dining rewards?"
python3 usecases/llm-banking-chatbot/app/banking_chatbot.py --offline-fixture --customer C003 --simulate-realtime-tx