diff --git a/README.md b/README.md
index 42b771b..1dae574 100644
--- a/README.md
+++ b/README.md
@@ -47,6 +47,7 @@ Refer to the [Installation section](https://getdozer.io/docs/installation) for i
 |                  | [IMDB Analytics](./usecases/imdb-analytics)                              | Use Dozer to get interesting analytics using an IMDb dataset                 |
 |                  | 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                      |
+|                  | [LLM Vector Banking Chatbot](./usecases/llm-vector-banking-chatbot)      | Dozer + LangChain + vector retrieval for card personalization (#1690)     |
 |                  |                                                                          |                                                                              |
 | Client Libraries | [Dozer React Starter](./usecases/react/)                                 | Instantly start building real time views using Dozer and React               |
 |                  | [Ingest Polars/Pandas Dataframes](./client-samples/ingest-python-sample) | Instantly ingest Polars/Pandas dataframes using Arrow format and deploy APIs |
diff --git a/usecases/llm-vector-banking-chatbot/.gitignore b/usecases/llm-vector-banking-chatbot/.gitignore
new file mode 100644
index 0000000..a5f2b9f
--- /dev/null
+++ b/usecases/llm-vector-banking-chatbot/.gitignore
@@ -0,0 +1,9 @@
+.dozer/
+__pycache__/
+*.pyc
+.venv/
+venv/
+.env
+.chroma/
+*.egg-info/
+.pytest_cache/
diff --git a/usecases/llm-vector-banking-chatbot/README.md b/usecases/llm-vector-banking-chatbot/README.md
new file mode 100644
index 0000000..2dc5716
--- /dev/null
+++ b/usecases/llm-vector-banking-chatbot/README.md
@@ -0,0 +1,105 @@
+# Dozer + LLM + Vector DB + LangChain (Banking Chatbot)
+
+Sample for [getdozer/dozer#1690](https://github.com/getdozer/dozer/issues/1690) ($250 Algora).
+
+Inspired by the Dozer hyper-personalization banking chatbot write-up: unify customer
+profiles, accounts, and transactions in Dozer, then pass that context into a LangChain
+retrieval flow over credit-card products (vector store + optional LLM).
+
+## What this sample includes
+
+| Piece | Path | Role |
+| --- | --- | --- |
+| Dozer config | `dozer-config.yaml` | LocalStorage CSV sources + spend-profile SQL + REST endpoints |
+| Seed data | `data/*/*.csv` | customers, accounts, transactions, card products |
+| RAG stub | `app/rag_chatbot.py` | fixture mode (offline) + `--use-dozer` + optional `--llm` |
+| Unit tests | `tests/test_rag_chatbot.py` | fixture load, segment gate, keyword/rerank |
+| Deps | `app/requirements.txt` | LangChain / Chroma / OpenAI (optional) |
+| Demo log | `demo-log.txt` | text capture of live Dozer REST + chatbot (no video yet) |
+
+## Prerequisites
+
+- **Dozer 0.3.x** on `PATH` for the live path (API-era binary with `endpoints`).
+  Dozer **≥0.4.0** removed built-in `endpoints` / `dozer-api` — that binary will reject
+  this config. See [`SAMPLE_NOTES.md`](./SAMPLE_NOTES.md).
+- Python 3.10+
+- `protoc` (`protobuf-compiler`) for `dozer build` on 0.3.x
+- Optional: `OPENAI_API_KEY` for the ChatOpenAI answer path
+
+## Quick start (offline fixture — no Dozer binary)
+
+```bash
+cd usecases/llm-vector-banking-chatbot
+python3 app/rag_chatbot.py \
+  --customer C001 \
+  --question "Which card is best for travel and dining rewards?"
+```
+
+Expected shape: a recommendation for **Dozer Travel Signature** grounded in C001's
+travel/dining spend.
+
+Student example:
+
+```bash
+python3 app/rag_chatbot.py \
+  --customer C003 \
+  --question "I need a no-fee student card for education"
+```
+
+## Live Dozer path (verified on 0.3.0)
+
+```bash
+cd usecases/llm-vector-banking-chatbot
+rm -rf .dozer
+dozer --ignore-pipe build -c dozer-config.yaml
+dozer --ignore-pipe run -c dozer-config.yaml
+# elsewhere:
+python3 -m venv .venv && .venv/bin/pip install -r app/requirements.txt
+.venv/bin/python app/rag_chatbot.py --use-dozer --customer C001 \
+  --question "Which card is best for travel and dining rewards?"
+```
+
+Dozer REST listens at `http://localhost:8080` (override with `--dozer-base` or
+`DOZER_API_BASE`). Responses are bare JSON lists; the client filters by `customer_id`.
+
+Optional LLM polish:
+
+```bash
+export OPENAI_API_KEY=sk-...
+.venv/bin/python app/rag_chatbot.py --use-dozer --llm --customer C001 \
+  --question "Which card is best for travel and dining rewards?"
+```
+
+## Tests
+
+```bash
+cd usecases/llm-vector-banking-chatbot
+python3 -m venv .venv && .venv/bin/pip install pytest httpx
+.venv/bin/pytest -q tests/test_rag_chatbot.py
+```
+
+## APIs exposed (once Dozer 0.3.x is up)
+
+- `GET /customer_profiles`
+- `GET /accounts`
+- `GET /transactions`
+- `GET /card_products`
+- `GET /customer_spending_profile` — SQL aggregate used as chatbot context
+
+## Architecture (sample)
+
+```
+CSV seeds ──► Dozer LocalStorage ──► SQL spend profile ──► REST/gRPC APIs
+                                                                  │
+                     LangChain retriever (Chroma / FakeEmbeddings) │
+                     + customer-aware rerank ◄─────────────────────┘
+                                                                  │
+                     deterministic answer  or  ChatOpenAI (--llm)
+```
+
+## Status / honesty
+
+See [`SAMPLE_NOTES.md`](./SAMPLE_NOTES.md). Live `dozer run` (0.3.0) + `--use-dozer`
+chatbot verified; text demo log attached. **No payout claimed.** Demo *video* and
+maintainer confirmation of bounty eligibility remain open. Crowded claim field
+(`dozer-samples` #139–#145, etc.).
diff --git a/usecases/llm-vector-banking-chatbot/SAMPLE_NOTES.md b/usecases/llm-vector-banking-chatbot/SAMPLE_NOTES.md
new file mode 100644
index 0000000..815207b
--- /dev/null
+++ b/usecases/llm-vector-banking-chatbot/SAMPLE_NOTES.md
@@ -0,0 +1,62 @@
+# SAMPLE_NOTES — getdozer/dozer#1690
+
+**Branch:** `bounty/1690-llm-vector-banking-scaffold`  
+**PR:** https://github.com/getdozer/dozer-samples/pull/144  
+**Issue:** https://github.com/getdozer/dozer/issues/1690  
+**Bounty:** $250 Algora — Dozer + LLM + Vector DB + LangChain sample  
+
+## Do NOT invent payout success
+
+No maintainer acceptance, no Algora payout confirmation. Multiple competing open PRs
+(`dozer-samples` #139–#145 and others). Algora may already show claims/completed —
+treat reward eligibility as **uncertain**.
+
+## What works now (verified locally)
+
+| Check | Result |
+| --- | --- |
+| Offline fixture chatbot | OK (C001→Travel Signature, C003→Campus Starter) |
+| `dozer build/run` **0.3.0** against `dozer-config.yaml` | OK (`--ignore-pipe`; needs `protoc` + OpenSSL 1.1) |
+| REST `GET /customer_profiles`, `/customer_spending_profile`, `/card_products` | 200, bare JSON lists |
+| `rag_chatbot.py --use-dozer` | OK after client-side customer filter fix |
+| Unit tests `tests/test_rag_chatbot.py` | 7 passed |
+| Text demo log | `demo-log.txt` |
+| Demo **video** | **Still missing** |
+
+### Commands that succeeded
+
+```text
+dozer --ignore-pipe build -c dozer-config.yaml   # dozer 0.3.0
+dozer --ignore-pipe run -c dozer-config.yaml
+python3 app/rag_chatbot.py --use-dozer --customer C001 --question "Which card is best for travel and dining rewards?"
+python3 app/rag_chatbot.py --use-dozer --customer C003 --question "I need a no-fee student card for education"
+pytest -q tests/test_rag_chatbot.py               # 7 passed
+```
+
+## Config / client fixes in this follow-up
+
+1. Explicit `api.rest` / `api.grpc` host+port (8080 / 50051).
+2. Composite PK on `customer_spending_profile`: `customer_id` + `top_category`
+   (matches `GROUP BY customer_id, category`).
+3. `--use-dozer` no longer trusts `?customer_id=` (0.3 ignores it and would always
+   pick the first profile); always filter client-side on bare lists.
+4. Document that **Dozer ≥0.4.0 rejects `endpoints`** (`unknown field endpoints`;
+   sinks-only era). This sample targets the API-era 0.3.x binary that matches the
+   historical LocalStorage samples in this repo.
+
+## Remaining gaps
+
+1. **No demo video** (Algora often wants ≤30s screen capture).
+2. **Architecture drift:** issue comments note current main may prefer
+   Dozer→sink→app API. This PR intentionally stays on 0.3-style REST endpoints to
+   match existing `dozer-samples` patterns; maintainers may ask for a sink rewrite.
+3. **Blog URL** `https://getdozer.io/blog/llm-chatbot` still 404 when last checked.
+4. **Crowded claims** — many open PRs; confirm with maintainers before banking on $250.
+5. Optional Chroma/LangChain path needs `pip install -r app/requirements.txt`; tests
+   and live demo use keyword retrieve + fixture/Dozer context without Chroma.
+
+## Confidence
+
+**~0.70** that this is a credible, runnable sample PR after live verify + tests.  
+**~0.20** that it wins $250 given competition + possible architecture preference for
+sinks — **do not bank on payout**.
diff --git a/usecases/llm-vector-banking-chatbot/app/__init__.py b/usecases/llm-vector-banking-chatbot/app/__init__.py
new file mode 100644
index 0000000..c0bf531
--- /dev/null
+++ b/usecases/llm-vector-banking-chatbot/app/__init__.py
@@ -0,0 +1 @@
+"""LangChain RAG stub for the Dozer banking chatbot sample."""
diff --git a/usecases/llm-vector-banking-chatbot/app/rag_chatbot.py b/usecases/llm-vector-banking-chatbot/app/rag_chatbot.py
new file mode 100644
index 0000000..3a237b6
--- /dev/null
+++ b/usecases/llm-vector-banking-chatbot/app/rag_chatbot.py
@@ -0,0 +1,276 @@
+#!/usr/bin/env python3
+"""
+Dozer + LangChain RAG stub for credit-card personalization.
+
+Modes
+-----
+* fixture (default): reads local CSVs — no Dozer binary, no API key required
+* --use-dozer: fetch customer context from Dozer REST (http://localhost:8080)
+* --llm: optional ChatOpenAI final answer when OPENAI_API_KEY is set
+
+Fixture mode needs no Dozer binary. Live mode (--use-dozer) was verified
+against Dozer 0.3.0 REST (bare JSON lists; client-side customer filter).
+"""
+
+from __future__ import annotations
+
+import argparse
+import csv
+import os
+import sys
+from collections import defaultdict
+from dataclasses import dataclass
+from pathlib import Path
+
+SAMPLE_ROOT = Path(__file__).resolve().parents[1]
+DATA_DIR = SAMPLE_ROOT / "data"
+DEFAULT_DOZER = os.environ.get("DOZER_API_BASE", "http://localhost:8080")
+
+
+@dataclass
+class CustomerContext:
+    customer_id: str
+    full_name: str
+    segment: str
+    city: str
+    spend_by_category: dict[str, float]
+    total_spend: float
+
+    def summary(self) -> str:
+        top = sorted(self.spend_by_category.items(), key=lambda x: -x[1])[:3]
+        top_s = ", ".join(f"{c}=${v:,.0f}" for c, v in top) or "n/a"
+        return (
+            f"{self.full_name} ({self.customer_id}), segment={self.segment}, "
+            f"city={self.city}, total_spend=${self.total_spend:,.0f}, top=[{top_s}]"
+        )
+
+
+def _csv_rows(table: str) -> list[dict[str, str]]:
+    path = DATA_DIR / table / f"{table}.csv"
+    with path.open(newline="", encoding="utf-8") as fh:
+        return list(csv.DictReader(fh))
+
+
+def load_fixture_context(customer_id: str) -> CustomerContext:
+    profiles = {r["customer_id"]: r for r in _csv_rows("customer_profiles")}
+    if customer_id not in profiles:
+        raise SystemExit(f"Unknown customer_id={customer_id}")
+    p = profiles[customer_id]
+    spend: dict[str, float] = defaultdict(float)
+    for txn in _csv_rows("transactions"):
+        if txn["customer_id"] == customer_id:
+            spend[txn["category"]] += float(txn["amount"])
+    return CustomerContext(
+        customer_id=customer_id,
+        full_name=p["full_name"],
+        segment=p["segment"],
+        city=p["city"],
+        spend_by_category=dict(spend),
+        total_spend=sum(spend.values()),
+    )
+
+
+def load_dozer_context(customer_id: str, base: str) -> CustomerContext:
+    """Fetch unified profile from Dozer REST once the binary is running.
+
+    Dozer 0.3 REST returns a bare JSON list and does not honor simple
+    ``?customer_id=`` query filters, so we always filter client-side.
+    """
+    try:
+        import httpx
+    except ImportError as exc:
+        raise SystemExit("httpx required for --use-dozer; pip install httpx") from exc
+
+    def get(path: str) -> list[dict]:
+        r = httpx.get(f"{base.rstrip('/')}{path}", timeout=10.0)
+        r.raise_for_status()
+        payload = r.json()
+        # Dozer 0.3 returns a bare list; older docs mentioned {"data": [...]}
+        if isinstance(payload, dict):
+            return payload.get("data") or payload.get("records") or []
+        if isinstance(payload, list):
+            return payload
+        return []
+
+    profiles = [p for p in get("/customer_profiles") if p.get("customer_id") == customer_id]
+    if not profiles:
+        raise SystemExit(f"Dozer returned no profile for {customer_id}")
+    p = profiles[0]
+    spend_rows = [
+        row
+        for row in get("/customer_spending_profile")
+        if row.get("customer_id") == customer_id
+    ]
+    spend = {row["top_category"]: float(row["total_spend"]) for row in spend_rows}
+    return CustomerContext(
+        customer_id=customer_id,
+        full_name=p.get("full_name", customer_id),
+        segment=p.get("segment", "unknown"),
+        city=p.get("city", ""),
+        spend_by_category=spend,
+        total_spend=sum(spend.values()),
+    )
+
+
+def load_card_docs() -> list[dict[str, str]]:
+    return _csv_rows("card_products")
+
+
+def segment_ok(customer_segment: str, min_segment: str) -> bool:
+    order = ["student", "mass_market", "mass_affluent", "premium"]
+    try:
+        return order.index(customer_segment) >= order.index(min_segment)
+    except ValueError:
+        return True
+
+
+def keyword_retrieve(question: str, cards: list[dict[str, str]], k: int = 3) -> list[dict[str, str]]:
+    q = question.lower()
+    scored: list[tuple[int, dict[str, str]]] = []
+    for card in cards:
+        blob = " ".join(
+            [
+                card["name"],
+                card["reward_categories"],
+                card["perk_summary"],
+                card["tags"],
+            ]
+        ).lower()
+        score = sum(1 for token in q.replace("?", " ").split() if token and token in blob)
+        scored.append((score, card))
+    scored.sort(key=lambda x: -x[0])
+    return [c for s, c in scored[:k] if s > 0] or [c for _, c in scored[:k]]
+
+
+def vector_retrieve(question: str, cards: list[dict[str, str]], k: int = 3) -> list[dict[str, str]]:
+    """LangChain + Chroma path when deps are installed; else keyword fallback."""
+    try:
+        from langchain_community.embeddings import FakeEmbeddings
+        from langchain_community.vectorstores import Chroma
+        from langchain_core.documents import Document
+    except Exception:
+        return keyword_retrieve(question, cards, k=k)
+
+    docs = [
+        Document(
+            page_content=(
+                f"{c['name']}. Rewards: {c['reward_categories']}. "
+                f"{c['perk_summary']}. Tags: {c['tags']}"
+            ),
+            metadata={"product_id": c["product_id"], "name": c["name"]},
+        )
+        for c in cards
+    ]
+    store = Chroma.from_documents(docs, embedding=FakeEmbeddings(size=32))
+    hits = store.similarity_search(question, k=k)
+    by_id = {c["product_id"]: c for c in cards}
+    out = []
+    for h in hits:
+        pid = h.metadata.get("product_id")
+        if pid in by_id:
+            out.append(by_id[pid])
+    return out or keyword_retrieve(question, cards, k=k)
+
+
+def rerank_for_customer(
+    ctx: CustomerContext,
+    candidates: list[dict[str, str]],
+    question: str = "",
+) -> list[dict[str, str]]:
+    top_cats = {c for c, _ in sorted(ctx.spend_by_category.items(), key=lambda x: -x[1])[:3]}
+    q_tokens = {t for t in question.lower().replace("?", " ").split() if len(t) > 2}
+
+    def score(card: dict[str, str]) -> float:
+        if not segment_ok(ctx.segment, card["min_segment"]):
+            return -1.0
+        rewards = set(card["reward_categories"].split("|"))
+        overlap = float(len(top_cats & rewards))
+        blob = f"{card['reward_categories']}|{card['tags']}|{card['name']}".lower()
+        q_hit = sum(1.0 for t in q_tokens if t in blob)
+        fee = float(card["annual_fee"])
+        fee_penalty = 1.0 if fee >= 100000 else (0.2 if fee >= 40000 else 0.0)
+        return overlap + 0.75 * q_hit - fee_penalty
+
+    ranked = sorted(candidates, key=score, reverse=True)
+    return [c for c in ranked if score(c) >= 0] or ranked
+
+
+def deterministic_answer(ctx: CustomerContext, question: str, picks: list[dict[str, str]]) -> str:
+    if not picks:
+        return (
+            f"No eligible card for {ctx.full_name} given segment={ctx.segment}. "
+            "Ask about everyday cashback or student products."
+        )
+    best = picks[0]
+    return (
+        f"For {ctx.summary()}, given question «{question}», "
+        f"recommend **{best['name']}** ({best['product_id']}): {best['perk_summary']} "
+        f"(annual fee {best['annual_fee']} ARS)."
+    )
+
+
+def maybe_llm_answer(ctx: CustomerContext, question: str, picks: list[dict[str, str]]) -> str | None:
+    if not os.environ.get("OPENAI_API_KEY"):
+        return None
+    try:
+        from langchain_openai import ChatOpenAI
+        from langchain_core.messages import HumanMessage, SystemMessage
+    except Exception:
+        return None
+
+    catalog = "\n".join(
+        f"- {c['name']}: {c['perk_summary']} (fee={c['annual_fee']})" for c in picks
+    )
+    llm = ChatOpenAI(model=os.environ.get("OPENAI_MODEL", "gpt-4o-mini"), temperature=0)
+    msg = llm.invoke(
+        [
+            SystemMessage(
+                content=(
+                    "You are a bank product assistant. Use only the provided customer "
+                    "context and card candidates. Be concise."
+                )
+            ),
+            HumanMessage(
+                content=(
+                    f"Customer context:\n{ctx.summary()}\n\n"
+                    f"Candidates:\n{catalog}\n\nQuestion: {question}"
+                )
+            ),
+        ]
+    )
+    return str(msg.content)
+
+
+def run(customer_id: str, question: str, use_dozer: bool, use_llm: bool, dozer_base: str) -> str:
+    ctx = (
+        load_dozer_context(customer_id, dozer_base)
+        if use_dozer
+        else load_fixture_context(customer_id)
+    )
+    cards = load_card_docs()
+    candidates = vector_retrieve(question, cards, k=4)
+    picks = rerank_for_customer(ctx, candidates, question=question)
+    if use_llm:
+        llm_text = maybe_llm_answer(ctx, question, picks)
+        if llm_text:
+            return llm_text
+    return deterministic_answer(ctx, question, picks)
+
+
+def main(argv: list[str] | None = None) -> int:
+    parser = argparse.ArgumentParser(description=__doc__)
+    parser.add_argument("--customer", default="C001")
+    parser.add_argument(
+        "--question",
+        default="Which card is best for travel and dining rewards?",
+    )
+    parser.add_argument("--use-dozer", action="store_true")
+    parser.add_argument("--llm", action="store_true", help="Try ChatOpenAI if key present")
+    parser.add_argument("--dozer-base", default=DEFAULT_DOZER)
+    args = parser.parse_args(argv)
+    print(run(args.customer, args.question, args.use_dozer, args.llm, args.dozer_base))
+    return 0
+
+
+if __name__ == "__main__":
+    sys.exit(main())
diff --git a/usecases/llm-vector-banking-chatbot/app/requirements.txt b/usecases/llm-vector-banking-chatbot/app/requirements.txt
new file mode 100644
index 0000000..392fffc
--- /dev/null
+++ b/usecases/llm-vector-banking-chatbot/app/requirements.txt
@@ -0,0 +1,7 @@
+# Optional live path (install when running against Dozer + OpenAI)
+langchain>=0.2.0
+langchain-community>=0.2.0
+langchain-openai>=0.1.0
+chromadb>=0.5.0
+httpx>=0.27.0
+openai>=1.30.0
diff --git a/usecases/llm-vector-banking-chatbot/data/accounts/accounts.csv b/usecases/llm-vector-banking-chatbot/data/accounts/accounts.csv
new file mode 100644
index 0000000..5f93ecc
--- /dev/null
+++ b/usecases/llm-vector-banking-chatbot/data/accounts/accounts.csv
@@ -0,0 +1,9 @@
+account_id,customer_id,product_type,currency,balance,status,opened_at
+A1001,C001,checking,ARS,845000.50,active,2019-03-12
+A1002,C001,credit_card,ARS,120000.00,active,2020-01-15
+A1003,C002,checking,ARS,210400.00,active,2021-07-01
+A1004,C002,savings,ARS,550000.00,active,2021-08-10
+A1005,C003,checking,ARS,45200.75,active,2023-01-18
+A1006,C004,checking,ARS,1500000.00,active,2018-11-05
+A1007,C004,credit_card,ARS,0.00,active,2019-02-01
+A1008,C005,checking,ARS,12800.00,active,2022-09-22
diff --git a/usecases/llm-vector-banking-chatbot/data/card_products/card_products.csv b/usecases/llm-vector-banking-chatbot/data/card_products/card_products.csv
new file mode 100644
index 0000000..43751ad
--- /dev/null
+++ b/usecases/llm-vector-banking-chatbot/data/card_products/card_products.csv
@@ -0,0 +1,6 @@
+product_id,name,annual_fee,reward_categories,perk_summary,min_segment,tags
+P_TRAVEL,Dozer Travel Signature,45000,"travel|dining","3x points on travel and dining, lounge access, travel insurance",premium,"travel,dining,lounge"
+P_CASHBACK,Dozer Everyday Cashback,0,"groceries|fuel|shopping","2% cash back on groceries and fuel, 1% everywhere else",mass_market,"cashback,everyday"
+P_STUDENT,Dozer Campus Starter,0,"education|dining|transit","1.5x points on education and transit, no annual fee",student,"student,education,no_fee"
+P_LUXURY,Dozer Black Reserve,120000,"shopping|travel","5x luxury shopping, concierge, airport transfer credits",premium,"luxury,concierge,high_fee"
+P_BALANCE,Dozer Balance Transfer,15000,"shopping","0% APR intro on transfers, modest rewards elsewhere",mass_affluent,"balance_transfer,apr"
diff --git a/usecases/llm-vector-banking-chatbot/data/customer_profiles/customer_profiles.csv b/usecases/llm-vector-banking-chatbot/data/customer_profiles/customer_profiles.csv
new file mode 100644
index 0000000..b3a3fa9
--- /dev/null
+++ b/usecases/llm-vector-banking-chatbot/data/customer_profiles/customer_profiles.csv
@@ -0,0 +1,6 @@
+customer_id,full_name,segment,city,preferred_channel,risk_tier,joined_at
+C001,Ava Chen,premium,Buenos Aires,mobile,low,2019-03-12
+C002,Mateo Ruiz,mass_affluent,Córdoba,web,medium,2021-07-01
+C003,Sofia Almeida,student,Rosario,mobile,medium,2023-01-18
+C004,Diego Morales,premium,Mendoza,branch,low,2018-11-05
+C005,Luna Ferreira,mass_market,La Plata,mobile,high,2022-09-22
diff --git a/usecases/llm-vector-banking-chatbot/data/transactions/transactions.csv b/usecases/llm-vector-banking-chatbot/data/transactions/transactions.csv
new file mode 100644
index 0000000..d623d21
--- /dev/null
+++ b/usecases/llm-vector-banking-chatbot/data/transactions/transactions.csv
@@ -0,0 +1,19 @@
+txn_id,account_id,customer_id,posted_at,merchant,category,amount,currency
+T9001,A1002,C001,2026-08-01,Aerolineas Argentinas,travel,185000.00,ARS
+T9002,A1002,C001,2026-08-03,Osaka Sushi,dining,42000.00,ARS
+T9003,A1002,C001,2026-08-05,Hilton Buenos Aires,travel,98000.00,ARS
+T9004,A1002,C001,2026-08-08,Mercado Libre,shopping,31000.00,ARS
+T9005,A1002,C001,2026-08-12,Cafe Tortoni,dining,8500.00,ARS
+T9006,A1003,C002,2026-08-02,Shell,fuel,28000.00,ARS
+T9007,A1003,C002,2026-08-04,Carrefour,groceries,45000.00,ARS
+T9008,A1003,C002,2026-08-09,Farmacity,health,12000.00,ARS
+T9009,A1003,C002,2026-08-14,Netflix,subscriptions,6500.00,ARS
+T9010,A1005,C003,2026-08-03,Udemy,education,22000.00,ARS
+T9011,A1005,C003,2026-08-07,Starbucks,dining,4800.00,ARS
+T9012,A1005,C003,2026-08-11,Subte BA,transit,2500.00,ARS
+T9013,A1007,C004,2026-08-01,Louis Vuitton,shopping,320000.00,ARS
+T9014,A1007,C004,2026-08-06,Four Seasons,travel,210000.00,ARS
+T9015,A1007,C004,2026-08-10,Tegui,dining,95000.00,ARS
+T9016,A1008,C005,2026-08-02,Easy,home,18000.00,ARS
+T9017,A1008,C005,2026-08-08,YPF,fuel,15000.00,ARS
+T9018,A1008,C005,2026-08-13,Coto,groceries,22000.00,ARS
diff --git a/usecases/llm-vector-banking-chatbot/demo-fixture.mp4 b/usecases/llm-vector-banking-chatbot/demo-fixture.mp4
new file mode 100644
index 0000000..ec1c95f
Binary files /dev/null and b/usecases/llm-vector-banking-chatbot/demo-fixture.mp4 differ
diff --git a/usecases/llm-vector-banking-chatbot/demo-log.txt b/usecases/llm-vector-banking-chatbot/demo-log.txt
new file mode 100644
index 0000000..1695718
--- /dev/null
+++ b/usecases/llm-vector-banking-chatbot/demo-log.txt
@@ -0,0 +1,61 @@
+=== DEMO LOG — llm-vector-banking-chatbot (Dozer 0.3.0 + RAG stub) ===
+Captured: 2026-09-08 05:46:36 ART
+
+--- dozer -V ---
+dozer 0.3.0
+
+--- REST endpoints (Dozer already running) ---
+["/customer_profiles","/accounts","/transactions","/card_products","/customer_spending_profile"]
+
+--- GET /customer_profiles (first record) ---
+{
+  "customer_id": "C001",
+  "full_name": "Ava Chen",
+  "segment": "premium",
+  "city": "Buenos Aires",
+  "preferred_channel": "mobile",
+  "risk_tier": "low",
+  "joined_at": "2019-03-12",
+  "__dozer_record_id": 0,
+  "__dozer_record_version": 1
+}
+
+--- GET /customer_spending_profile (C001 rows) ---
+[
+  {
+    "customer_id": "C001",
+    "txn_count": 2,
+    "total_spend": 283000.0,
+    "top_category": "travel",
+    "__dozer_record_id": 0,
+    "__dozer_record_version": 2
+  },
+  {
+    "customer_id": "C001",
+    "txn_count": 1,
+    "total_spend": 31000.0,
+    "top_category": "shopping",
+    "__dozer_record_id": 2,
+    "__dozer_record_version": 1
+  },
+  {
+    "customer_id": "C001",
+    "txn_count": 2,
+    "total_spend": 50500.0,
+    "top_category": "dining",
+    "__dozer_record_id": 1,
+    "__dozer_record_version": 2
+  }
+]
+
+--- chatbot --use-dozer C001 ---
+For Ava Chen (C001), segment=premium, city=Buenos Aires, total_spend=$364,500, top=[travel=$283,000, dining=$50,500, shopping=$31,000], given question «Which card is best for travel and dining rewards?», recommend **Dozer Travel Signature** (P_TRAVEL): 3x points on travel and dining, lounge access, travel insurance (annual fee 45000 ARS).
+
+--- chatbot --use-dozer C003 ---
+For Sofia Almeida (C003), segment=student, city=Rosario, total_spend=$29,300, top=[education=$22,000, dining=$4,800, transit=$2,500], given question «I need a no-fee student card for education», recommend **Dozer Campus Starter** (P_STUDENT): 1.5x points on education and transit, no annual fee (annual fee 0 ARS).
+
+--- pytest ---
+.......                                                                  [100%]
+7 passed in 0.01s
+
+=== END DEMO LOG ===
diff --git a/usecases/llm-vector-banking-chatbot/dozer-config.yaml b/usecases/llm-vector-banking-chatbot/dozer-config.yaml
new file mode 100644
index 0000000..c5e4bd9
--- /dev/null
+++ b/usecases/llm-vector-banking-chatbot/dozer-config.yaml
@@ -0,0 +1,105 @@
+# Dozer + LLM banking chatbot sample (issue getdozer/dozer#1690)
+# Sources multiple LocalStorage CSV datasets, materializes a unified
+# customer spending profile, and exposes REST/gRPC APIs for LangChain.
+#
+# Verified against Dozer 0.3.0 (API-era binary). Dozer >=0.4.0 removed
+# built-in `endpoints` / dozer-api; see SAMPLE_NOTES.md.
+app_name: llm-vector-banking-chatbot
+version: 1
+
+api:
+  rest:
+    port: 8080
+    host: 0.0.0.0
+  grpc:
+    port: 50051
+    host: 0.0.0.0
+
+connections:
+  - name: bank_local
+    config: !LocalStorage
+      details:
+        path: data
+      tables:
+        - !Table
+          name: customer_profiles
+          config: !CSV
+            path: customer_profiles
+            extension: .csv
+        - !Table
+          name: accounts
+          config: !CSV
+            path: accounts
+            extension: .csv
+        - !Table
+          name: transactions
+          config: !CSV
+            path: transactions
+            extension: .csv
+        - !Table
+          name: card_products
+          config: !CSV
+            path: card_products
+            extension: .csv
+
+sources:
+  - name: customer_profiles
+    table_name: customer_profiles
+    connection: bank_local
+  - name: accounts
+    table_name: accounts
+    connection: bank_local
+  - name: transactions
+    table_name: transactions
+    connection: bank_local
+  - name: card_products
+    table_name: card_products
+    connection: bank_local
+
+sql: |
+  -- Unified spend profile: total spend + category per customer
+  SELECT
+    t.customer_id AS customer_id,
+    COUNT(t.txn_id) AS txn_count,
+    SUM(t.amount) AS total_spend,
+    t.category AS top_category
+  INTO customer_spending_profile
+  FROM transactions t
+  GROUP BY t.customer_id, t.category;
+
+endpoints:
+  - name: customer_profiles
+    path: /customer_profiles
+    table_name: customer_profiles
+    index:
+      primary_key:
+        - customer_id
+
+  - name: accounts
+    path: /accounts
+    table_name: accounts
+    index:
+      primary_key:
+        - account_id
+
+  - name: transactions
+    path: /transactions
+    table_name: transactions
+    index:
+      primary_key:
+        - txn_id
+
+  - name: card_products
+    path: /card_products
+    table_name: card_products
+    index:
+      primary_key:
+        - product_id
+
+  - name: customer_spending_profile
+    path: /customer_spending_profile
+    table_name: customer_spending_profile
+    index:
+      primary_key:
+        - customer_id
+        - top_category
diff --git a/usecases/llm-vector-banking-chatbot/tests/test_rag_chatbot.py b/usecases/llm-vector-banking-chatbot/tests/test_rag_chatbot.py
new file mode 100644
index 0000000..1fadea4
--- /dev/null
+++ b/usecases/llm-vector-banking-chatbot/tests/test_rag_chatbot.py
@@ -0,0 +1,80 @@
+"""Unit tests for the offline fixture / ranking path (no Dozer binary required)."""
+
+from __future__ import annotations
+
+import sys
+from pathlib import Path
+
+import pytest
+
+ROOT = Path(__file__).resolve().parents[1]
+sys.path.insert(0, str(ROOT / "app"))
+
+from rag_chatbot import (  # noqa: E402
+    CustomerContext,
+    deterministic_answer,
+    keyword_retrieve,
+    load_card_docs,
+    load_fixture_context,
+    rerank_for_customer,
+    segment_ok,
+)
+
+
+def test_segment_ok_ordering() -> None:
+    assert segment_ok("premium", "student")
+    assert segment_ok("student", "student")
+    assert not segment_ok("student", "premium")
+    assert segment_ok("mass_affluent", "mass_market")
+
+
+def test_load_fixture_context_c001() -> None:
+    ctx = load_fixture_context("C001")
+    assert ctx.full_name == "Ava Chen"
+    assert ctx.segment == "premium"
+    assert ctx.spend_by_category["travel"] == pytest.approx(283000.0)
+    assert ctx.total_spend == pytest.approx(364500.0)
+
+
+def test_load_fixture_unknown_customer() -> None:
+    with pytest.raises(SystemExit):
+        load_fixture_context("NOPE")
+
+
+def test_keyword_retrieve_travel() -> None:
+    cards = load_card_docs()
+    hits = keyword_retrieve("travel dining lounge rewards", cards, k=3)
+    assert hits
+    assert hits[0]["product_id"] == "P_TRAVEL"
+
+
+def test_rerank_prefers_travel_for_premium_spender() -> None:
+    ctx = load_fixture_context("C001")
+    cards = load_card_docs()
+    ranked = rerank_for_customer(ctx, cards, question="travel and dining rewards")
+    assert ranked[0]["product_id"] == "P_TRAVEL"
+
+
+def test_rerank_gates_student_away_from_luxury() -> None:
+    ctx = load_fixture_context("C003")
+    cards = load_card_docs()
+    ranked = rerank_for_customer(ctx, cards, question="no-fee student education")
+    ids = [c["product_id"] for c in ranked]
+    assert "P_LUXURY" not in ids
+    assert "P_TRAVEL" not in ids
+    assert ranked[0]["product_id"] == "P_STUDENT"
+
+
+def test_deterministic_answer_mentions_card() -> None:
+    ctx = CustomerContext(
+        customer_id="C001",
+        full_name="Ava Chen",
+        segment="premium",
+        city="Buenos Aires",
+        spend_by_category={"travel": 100.0},
+        total_spend=100.0,
+    )
+    cards = load_card_docs()
+    text = deterministic_answer(ctx, "travel?", [cards[0]])
+    assert "Dozer Travel Signature" in text
+    assert "C001" in text