Add Dozer + LLM + vector database + LangChain sample

README.md

@@ -45,6 +45,7 @@ Refer to the [Installation section](https://getdozer.io/docs/installation) for i
 | Use Cases        | [Flight Microservices](./usecases/pg-flights)                            | Build APIs over multiple microservices.                                      |
 |                  | [Scaling Ecommerce](./usecases/scaling-ecommerce)                        | Profile and benchmark Dozer using an ecommerce data set                      |
 |                  | [IMDB Analytics](./usecases/imdb-analytics)                              | Use Dozer to get interesting analytics using an IMDb dataset                 |
+|                  | [LLM Chatbot](./usecases/llm-langchain)                                  | Recommend credit cards using Dozer, LangChain, and a vector database         |
 |                  | Use Dozer to Instrument (Coming soon)                                    | Combine Log data to get real time insights                                   |
 |                  | Real Time Model Scoring (Coming soon)                                    | Deploy trained models to get real time insights as APIs                      |
 |                  |                                                                          |                                                                              |

usecases/llm-langchain/README.md

@@ -0,0 +1,36 @@
+# Dozer + LLM + Vector Database + LangChain
+
+This sample implements the flow from the [LLM chatbot blog post](https://getdozer.io/blog/llm-chatbot): a bank chatbot that hyper-personalizes credit card recommendations.
+
+Dozer sources multiple datasets (customer profiles, transactions, card products) from local CSV files, builds a real-time spending aggregate in SQL, and serves it all as low-latency REST APIs. LangChain embeds the card products into a Chroma vector database and passes the unified customer profile as context to an LLM.
+
+## Prerequisites
+
+- [Dozer](https://getdozer.io/docs/installation)
+- Python 3.9 or higher
+- An OpenAI API key
+
+## Run
+
+### 1. Install the Python dependencies
+
+```bash
+pip install -r requirements.txt
+export OPENAI_API_KEY=sk-...
+```
+
+### 2. Start Dozer
+
+```bash
+dozer run -c dozer-config.yaml
+```
+
+Dozer exposes REST APIs on port 8080: `/customers`, `/transactions`, `/card_products`, and `/customer_spending` (per-customer spending aggregated by category).
+
+### 3. Chat
+
+```bash
+python chatbot.py --customer C001 --question "Which card best fits this customer, and why?"
+```
+
+The chatbot fetches the customer profile and spending breakdown from Dozer, retrieves the most relevant card products from the Chroma vector store, and asks the LLM for a personalized recommendation.

usecases/llm-langchain/chatbot.py

@@ -0,0 +1,97 @@
+"""Bank chatbot powered by Dozer REST APIs, LangChain, and a Chroma vector store.
+
+Dozer serves customers, transactions, and a spending aggregate as REST APIs.
+This script embeds card products into Chroma and lets an LLM recommend a card
+against the unified customer profile fetched from Dozer.
+"""
+
+import argparse
+import os
+
+import requests
+from langchain.chains import RetrievalQA
+from langchain_community.vectorstores import Chroma
+from langchain_openai import ChatOpenAI, OpenAIEmbeddings
+
+DOZER_API = os.environ.get("DOZER_API", "http://localhost:8080")
+
+
+def fetch_records(endpoint, limit=1000):
+    response = requests.get(
+        f"{DOZER_API}{endpoint}", params={"$limit": limit}, timeout=30
+    )
+    response.raise_for_status()
+    return response.json().get("records", [])
+
+
+def build_vector_store(products):
+    descriptions = [
+        (
+            f"{p['product_name']} (annual fee ${p['annual_fee']}, "
+            f"min income {p['min_income']}): {p['description']}"
+        )
+        for p in products
+    ]
+    return Chroma.from_texts(
+        descriptions,
+        embedding=OpenAIEmbeddings(),
+        collection_name="card_products",
+    )
+
+
+def customer_context(customers, spending, customer_id):
+    customer = next(
+        (c for c in customers if c["customer_id"] == customer_id), None
+    )
+    if customer is None:
+        raise SystemExit(
+            f"Customer {customer_id} not found at the /customers endpoint."
+        )
+    breakdown = [
+        f"{s['category']}: ${s['total_amount']}"
+        for s in spending
+        if s["customer_id"] == customer_id
+    ]
+    return (
+        f"{customer['name']}, age {customer['age']}, "
+        f"income ${customer['income']}, "
+        f"spending breakdown: {', '.join(breakdown) or 'no transactions'}"
+    )
+
+
+def main():
+    if not os.environ.get("OPENAI_API_KEY"):
+        raise SystemExit("Set OPENAI_API_KEY before running the chatbot.")
+
+    parser = argparse.ArgumentParser(description=__doc__)
+    parser.add_argument(
+        "--customer",
+        default="C001",
+        help="customer id from the Dozer /customers endpoint (default C001)",
+    )
+    parser.add_argument(
+        "--question",
+        default="Which card best fits this customer, and why?",
+        help="question to ask the LLM about the customer",
+    )
+    args = parser.parse_args()
+
+    customers = fetch_records("/customers")
+    spending = fetch_records("/customer_spending")
+    products = fetch_records("/card_products")
+
+    vector_store = build_vector_store(products)
+    chain = RetrievalQA.from_chain_type(
+        llm=ChatOpenAI(model="gpt-4o-mini", temperature=0),
+        retriever=vector_store.as_retriever(search_kwargs={"k": 2}),
+    )
+    prompt = (
+        f"Customer context: "
+        f"{customer_context(customers, spending, args.customer)}\n"
+        f"Question: {args.question}"
+    )
+    print(chain.invoke(prompt)["result"])
+
+
+if __name__ == "__main__":
+    main()

usecases/llm-langchain/data/card_products.csv

@@ -0,0 +1,5 @@
+card_id,product_name,description,annual_fee,min_income
+P001,Explorer Travel Card,2x points on travel and dining plus lounge access,95.0,40000
+P002,Cashback Everyday Card,3% cashback on grocery and fuel with no annual fee,0.0,0
+P003,Premium Rewards Card,5x points on retail and travel with concierge service,450.0,120000
+P004,Starter Student Card,Low credit limit card for students with no fees,0.0,20000

usecases/llm-langchain/data/customers.csv

@@ -0,0 +1,7 @@
+customer_id,name,age,income,city
+C001,Ava Patel,29,95000.0,New York
+C002,Marco Ruiz,41,62000.0,Austin
+C003,Priya Nair,35,145000.0,San Francisco
+C004,Dan Okafor,23,38000.0,Chicago
+C005,Lena Schmidt,52,98000.0,Seattle
+C006,Ryo Tanaka,47,73000.0,Boston

usecases/llm-langchain/data/transactions.csv

@@ -0,0 +1,16 @@
+transaction_id,customer_id,category,amount
+T001,C001,TRAVEL,820.5
+T002,C001,DINING,64.0
+T003,C002,GROCERY,120.3
+T004,C002,FUEL,45.8
+T005,C003,TRAVEL,1540.0
+T006,C003,DINING,215.4
+T007,C004,GROCERY,88.2
+T008,C004,ONLINE,31.9
+T009,C005,RETAIL,540.0
+T010,C005,TRAVEL,260.0
+T011,C006,GROCERY,96.4
+T012,C006,FUEL,52.1
+T013,C001,TRAVEL,310.0
+T014,C003,RETAIL,899.9
+T015,C002,GROCERY,140.6

usecases/llm-langchain/dozer-config.yaml

@@ -0,0 +1,71 @@
+app_name: llm-langchain
+version: 1
+connections:
+  - name: bank_data
+    config: !LocalStorage
+      details:
+        path: data
+      tables:
+        - !Table
+          name: customers
+          config: !Csv
+            path: customers
+            extension: .csv
+        - !Table
+          name: transactions
+          config: !Csv
+            path: transactions
+            extension: .csv
+        - !Table
+          name: card_products
+          config: !Csv
+            path: card_products
+            extension: .csv
+
+sql: |
+  SELECT
+    customer_id,
+    category,
+    SUM(amount) as total_amount,
+    COUNT(transaction_id) as transaction_count
+  INTO customer_spending
+  FROM transactions
+  GROUP BY customer_id, category;
+
+sources:
+  - name: customers
+    table_name: customers
+    connection: !Ref bank_data
+  - name: transactions
+    table_name: transactions
+    connection: !Ref bank_data
+  - name: card_products
+    table_name: card_products
+    connection: !Ref bank_data
+
+endpoints:
+  - name: customers
+    path: /customers
+    table_name: customers
+    index:
+      primary_key:
+        - customer_id
+  - name: transactions
+    path: /transactions
+    table_name: transactions
+    index:
+      primary_key:
+        - transaction_id
+  - name: card_products
+    path: /card_products
+    table_name: card_products
+    index:
+      primary_key:
+        - card_id
+  - name: customer_spending
+    path: /customer_spending
+    table_name: customer_spending
+    index:
+      primary_key:
+        - customer_id
+        - category

usecases/llm-langchain/requirements.txt

@@ -0,0 +1,5 @@
+langchain
+langchain-openai
+langchain-community
+chromadb
+requests