diff --git a/LSF-vLLM/README.md b/LSF-vLLM/README.md
new file mode 100644
index 0000000..fc2e3ae
--- /dev/null
+++ b/LSF-vLLM/README.md
@@ -0,0 +1,248 @@
+IBM LSF for vLLM Persistent Inference Service
+==========================================
+
+Overview
+--------
+This repository shows how to run a long-running vLLM inference service under IBM LSF,
+validate it through a standard OpenAI-compatible API, access it from a Jupyter notebook,
+and reuse the same service from a downstream batch job.
+
+What this implementation demonstrates
+-------------------------------------
+- IBM LSF launching and managing a persistent inference runtime as a service job
+- vLLM exposing an OpenAI-compatible endpoint
+- endpoint discovery through a small registry file written by the service job
+- interactive validation using curl and Jupyter
+- downstream reuse through a separate IBM LSF batch job
+
+Repository layout
+-----------------
+- scripts/start_vllm_lsf.sh
+  Starts the vLLM container, waits for readiness, writes the registry file, and keeps the
+  service attached to the IBM LSF job lifecycle.
+- scripts/resolve_endpoint.py
+  Reads the registry file for a given IBM LSF job ID and prints the resolved base URL.
+- scripts/batch_client.py
+  Reads a prompt corpus and sends requests to the registered vLLM service.
+- notebook/LSF_vLLM_Client.ipynb
+  Jupyter notebook for interactive validation against the IBM LSF-managed runtime.
+- corpus/prompts.txt
+  Sample prompt corpus for downstream batch validation.
+
+Prerequisites
+-------------
+- IBM LSF installed and operational
+- podman installed
+- python3 installed
+- curl installed
+- network access from the execution host to pull the vLLM image and model
+- a single-node IBM LSF setup is sufficient for this implementation
+
+Note:
+Replace "your-host" with the hostname or IP address of the system where the vLLM service is running.
+
+The examples below assume you are running as the same user for all steps.
+
+Step 1: Create the working directories
+--------------------------------------
+
+```bash
+mkdir -p ~/lsf_vllm_poc/{logs,registry,cache,corpus,results,notebook}
+```
+
+```bash
+cp corpus/prompts.txt ~/lsf_vllm_poc/corpus/prompts.txt
+```
+
+```bash
+cp scripts/start_vllm_lsf.sh ~/lsf_vllm_poc/
+cp scripts/resolve_endpoint.py ~/lsf_vllm_poc/
+cp scripts/batch_client.py ~/lsf_vllm_poc/
+```
+
+```bash
+chmod +x ~/lsf_vllm_poc/start_vllm_lsf.sh
+chmod +x ~/lsf_vllm_poc/resolve_endpoint.py
+chmod +x ~/lsf_vllm_poc/batch_client.py
+```
+
+Step 2: Review the service script defaults
+------------------------------------------
+
+```bash
+MODEL=Qwen/Qwen3-0.6B PORT=8001 API_KEY=local-vllm-key
+```
+
+Step 3: Submit the persistent service job
+-----------------------------------------
+
+```bash
+JOBID=$(
+  bsub -J vllm_service        -q normal        -n 1        -R 'rusage[mem=12GB]'        -oo ~/lsf_vllm_poc/logs/vllm.%J.out        -eo ~/lsf_vllm_poc/logs/vllm.%J.err        ~/lsf_vllm_poc/start_vllm_lsf.sh   | awk '{print $2}' | tr -d '<>'
+)
+
+echo "Submitted service JOBID=$JOBID"
+```
+
+Step 4: Monitor the service startup
+-----------------------------------
+
+```bash
+bjobs
+bjobs -l ${JOBID}
+bpeek ${JOBID}
+```
+
+```bash
+podman ps -a | grep vllm-job-${JOBID}
+podman logs -f vllm-job-${JOBID}
+```
+
+Step 5: Wait for the registry file
+----------------------------------
+
+```bash
+until [[ -f ~/lsf_vllm_poc/registry/${JOBID}.json ]]; do
+  sleep 2
+done
+
+cat ~/lsf_vllm_poc/registry/${JOBID}.json
+```
+
+Step 6: Resolve the endpoint
+----------------------------
+
+```bash
+python3 ~/lsf_vllm_poc/resolve_endpoint.py ${JOBID}
+```
+
+```bash
+ENDPOINT=$(python3 ~/lsf_vllm_poc/resolve_endpoint.py ${JOBID})
+echo "${ENDPOINT}"
+```
+
+Step 7: Validate the service with curl
+--------------------------------------
+
+```bash
+curl -sS "${ENDPOINT}/models"   -H "Authorization: Bearer local-vllm-key"
+```
+
+```bash
+curl -sS "${ENDPOINT}/chat/completions"   -H "Content-Type: application/json"   -H "Authorization: Bearer local-vllm-key"   -d '{
+    "model": "Qwen/Qwen3-0.6B",
+    "messages": [
+      {"role": "user", "content": "Explain the top 5 deserts in the world."}
+    ],
+    "temperature": 0,
+    "max_tokens": 120
+  }'
+```
+
+Step 8: Validate the service from Jupyter
+-----------------------------------------
+
+```bash
+python3 -m venv ~/lsf_vllm_poc/notebook/.venv
+source ~/lsf_vllm_poc/notebook/.venv/bin/activate
+pip install --upgrade pip
+pip install notebook jupyterlab requests openai ipykernel
+python -m ipykernel install --user --name lsf-vllm --display-name "Python (lsf-vllm)"
+```
+
+```bash
+jupyter notebook --no-browser --ip=0.0.0.0 --port 8888 --allow-root
+```
+
+```bash
+ssh -L 8888:127.0.0.1:8888 user@your-host
+```
+
+```
+http://127.0.0.1:8888
+```
+
+```
+http://127.0.0.1:8001/v1
+```
+
+Step 9: Validate downstream batch reuse
+---------------------------------------
+
+```bash
+python3 ~/lsf_vllm_poc/batch_client.py ${JOBID} ~/lsf_vllm_poc/corpus/prompts.txt
+```
+
+```bash
+cat ~/lsf_vllm_poc/results/batch_${JOBID}.jsonl
+```
+
+```bash
+BATCH_JOBID=$(
+  bsub -J vllm_batch        -q normal        -n 1        -R 'rusage[mem=1GB]'        -oo ~/lsf_vllm_poc/logs/batch.%J.out        -eo ~/lsf_vllm_poc/logs/batch.%J.err        "python3 ~/lsf_vllm_poc/batch_client.py ${JOBID} ~/lsf_vllm_poc/corpus/prompts.txt"   | awk '{print $2}' | tr -d '<>'
+)
+
+echo "Submitted batch JOBID=$BATCH_JOBID"
+```
+
+```bash
+bjobs
+bpeek ${BATCH_JOBID}
+cat ~/lsf_vllm_poc/results/batch_${JOBID}.jsonl
+```
+
+Cleanup
+-------
+
+```bash
+bkill ${BATCH_JOBID}
+bkill ${JOBID}
+```
+
+Troubleshooting
+---------------
+
+```bash
+bpeek ${JOBID}
+podman ps -a
+podman logs vllm-job-${JOBID}
+```
+
+```bash
+curl -sS http://127.0.0.1:8001/v1/models -H "Authorization: Bearer local-vllm-key"
+```
+
+Success criteria
+----------------
+1. the service job is RUN under IBM LSF
+2. the registry file exists
+3. /v1/models works
+4. /v1/chat/completions works
+5. the Jupyter notebook can call the service successfully
+6. the batch client works locally
+7. the IBM LSF batch job completes successfully
+8. the result file contains responses for the full prompt corpus
+
+Step 10: Validate using Open WebUI (Linux)
+------------------------------------------
+
+```bash
+podman run -d   -p 3000:8080   -e OPENAI_API_BASE_URL=http://your-host:8001/v1   -e OPENAI_API_KEY=local-vllm-key   -e WEBUI_SECRET_KEY=my-openwebui-secret   -v open-webui:/app/backend/data   --name open-webui   ghcr.io/open-webui/open-webui:main
+```
+
+```bash
+podman ps
+podman logs -f open-webui
+```
+
+http://localhost:3000
+
+- Settings → Connections → OpenAI
+- Base URL: http://your-host:8001/v1
+- API Key: local-vllm-key
+
+Model:
+Qwen/Qwen3-0.6B
+
+Test:
+Say one short line about LSF-managed model serving.
diff --git a/LSF-vLLM/scripts/LSF_vLLM_Client.ipynb b/LSF-vLLM/scripts/LSF_vLLM_Client.ipynb
new file mode 100644
index 0000000..21decf4
--- /dev/null
+++ b/LSF-vLLM/scripts/LSF_vLLM_Client.ipynb
@@ -0,0 +1,95 @@
+{
+  "cells": [
+    {
+      "cell_type": "markdown",
+      "metadata": {},
+      "source": [
+        "# IBM LSF + vLLM Notebook Validation\\n",
+        "Run cells top to bottom. Update values if your endpoint differs."
+      ]
+    },
+    {
+      "cell_type": "code",
+      "execution_count": null,
+      "metadata": {},
+      "outputs": [],
+      "source": [
+        "import requests\\n",
+        "base_url = 'http://127.0.0.1:8001/v1'\\n",
+        "api_key = 'local-vllm-key'\\n",
+        "model = 'Qwen/Qwen3-0.6B'\\n",
+        "print(base_url)\\n",
+        "print(model)\\n"
+      ]
+    },
+    {
+      "cell_type": "code",
+      "execution_count": null,
+      "metadata": {},
+      "outputs": [],
+      "source": [
+        "resp = requests.get(\\n",
+        "    f'{base_url}/models',\\n",
+        "    headers={'Authorization': f'Bearer {api_key}'},\\n",
+        "    timeout=60,\\n",
+        ")\\n",
+        "print(resp.status_code)\\n",
+        "print(resp.json())\\n"
+      ]
+    },
+    {
+      "cell_type": "code",
+      "execution_count": null,
+      "metadata": {},
+      "outputs": [],
+      "source": [
+        "payload = {\\n",
+        "    'model': model,\\n",
+        "    'messages': [{'role': 'user', 'content': 'Explain the top 5 deserts in the world.'}],\\n",
+        "    'temperature': 0,\\n",
+        "    'max_tokens': 120,\\n",
+        "    'chat_template_kwargs': {'enable_thinking': False},\\n",
+        "}\\n",
+        "resp = requests.post(\\n",
+        "    f'{base_url}/chat/completions',\\n",
+        "    headers={'Authorization': f'Bearer {api_key}', 'Content-Type': 'application/json'},\\n",
+        "    json=payload,\\n",
+        "    timeout=120,\\n",
+        ")\\n",
+        "print(resp.status_code)\\n",
+        "data = resp.json()\\n",
+        "print(data['choices'][0]['message']['content'])\\n"
+      ]
+    },
+    {
+      "cell_type": "code",
+      "execution_count": null,
+      "metadata": {},
+      "outputs": [],
+      "source": [
+        "from openai import OpenAI\\n",
+        "client = OpenAI(base_url=base_url, api_key=api_key)\\n",
+        "resp = client.chat.completions.create(\\n",
+        "    model=model,\\n",
+        "    messages=[{'role': 'user', 'content': 'Explain the top 5 deserts in the world.'}],\\n",
+        "    temperature=0,\\n",
+        "    max_tokens=120,\\n",
+        ")\\n",
+        "print(resp.choices[0].message.content)\\n"
+      ]
+    }
+  ],
+  "metadata": {
+    "kernelspec": {
+      "display_name": "Python (lsf-vllm)",
+      "language": "python",
+      "name": "python3"
+    },
+    "language_info": {
+      "name": "python",
+      "version": "3.9"
+    }
+  },
+  "nbformat": 4,
+  "nbformat_minor": 5
+}
\ No newline at end of file
diff --git a/LSF-vLLM/scripts/batch_client.py b/LSF-vLLM/scripts/batch_client.py
new file mode 100644
index 0000000..c386829
--- /dev/null
+++ b/LSF-vLLM/scripts/batch_client.py
@@ -0,0 +1,53 @@
+#!/usr/bin/env python3
+import json
+import os
+import sys
+import urllib.request
+
+if len(sys.argv) not in (2, 3):
+    print("Usage: batch_client.py <JOBID> [PROMPTS_FILE]", file=sys.stderr)
+    sys.exit(1)
+
+jobid = sys.argv[1]
+base = os.path.expanduser("~/lsf_vllm_poc")
+prompts_file = sys.argv[2] if len(sys.argv) == 3 else os.path.join(base, "corpus", "prompts.txt")
+
+with open(os.path.join(base, "registry", f"{jobid}.json")) as f:
+    reg = json.load(f)
+
+url = f"http://{reg['host']}:{reg['port']}/v1/chat/completions"
+headers = {
+    "Content-Type": "application/json",
+    "Authorization": f"Bearer {reg['api_key']}",
+}
+
+outp = os.path.join(base, "results", f"batch_{jobid}.jsonl")
+
+with open(prompts_file) as fin, open(outp, "w") as fout:
+    for line in fin:
+        prompt = line.strip()
+        if not prompt:
+            continue
+
+        payload = {
+            "model": reg["model"],
+            "messages": [{"role": "user", "content": prompt}],
+            "temperature": 0,
+            "max_tokens": 96,
+            "chat_template_kwargs": {"enable_thinking": False},
+        }
+
+        req = urllib.request.Request(
+            url,
+            data=json.dumps(payload).encode("utf-8"),
+            headers=headers,
+            method="POST",
+        )
+
+        with urllib.request.urlopen(req, timeout=300) as resp:
+            data = json.load(resp)
+
+        text = data["choices"][0]["message"]["content"]
+        fout.write(json.dumps({"prompt": prompt, "response": text}) + "\n")
+
+print(outp)
diff --git a/LSF-vLLM/scripts/prompts.txt b/LSF-vLLM/scripts/prompts.txt
new file mode 100644
index 0000000..1ce667a
--- /dev/null
+++ b/LSF-vLLM/scripts/prompts.txt
@@ -0,0 +1,3 @@
+Why do rivers flow towards the sea?
+What is the difference between a lion and a tiger?
+How do fish survive in water without breathing air like humans?
diff --git a/LSF-vLLM/scripts/resolve_endpoint.py b/LSF-vLLM/scripts/resolve_endpoint.py
new file mode 100644
index 0000000..cfb0f2e
--- /dev/null
+++ b/LSF-vLLM/scripts/resolve_endpoint.py
@@ -0,0 +1,16 @@
+#!/usr/bin/env python3
+import json
+import os
+import sys
+
+if len(sys.argv) != 2:
+    print("Usage: resolve_endpoint.py <JOBID>", file=sys.stderr)
+    sys.exit(1)
+
+jobid = sys.argv[1]
+reg_path = os.path.expanduser(f"~/lsf_vllm_poc/registry/{jobid}.json")
+
+with open(reg_path) as f:
+    reg = json.load(f)
+
+print(f"http://{reg['host']}:{reg['port']}/v1")
diff --git a/LSF-vLLM/scripts/start_vllm_lsf.sh b/LSF-vLLM/scripts/start_vllm_lsf.sh
new file mode 100644
index 0000000..d77b5b0
--- /dev/null
+++ b/LSF-vLLM/scripts/start_vllm_lsf.sh
@@ -0,0 +1,99 @@
+#!/usr/bin/env bash
+set -Eeuo pipefail
+
+MODEL="${MODEL:-Qwen/Qwen3-0.6B}"
+PORT="${PORT:-8001}"
+API_KEY="${API_KEY:-local-vllm-key}"
+IMAGE="${IMAGE:-docker.io/vllm/vllm-openai-cpu:latest-x86_64}"
+
+BASE="${HOME}/lsf_vllm_poc"
+CACHE_DIR="${BASE}/cache"
+REG_DIR="${BASE}/registry"
+LOG_DIR="${BASE}/logs"
+CONTAINER_NAME="vllm-job-${LSB_JOBID:-manual}"
+
+mkdir -p "${CACHE_DIR}" "${REG_DIR}" "${LOG_DIR}"
+
+echo "Starting vLLM service wrapper..."
+echo "MODEL=${MODEL}"
+echo "PORT=${PORT}"
+echo "JOBID=${LSB_JOBID:-manual}"
+echo "CONTAINER_NAME=${CONTAINER_NAME}"
+
+cleanup() {
+  echo "Cleaning up container ${CONTAINER_NAME}"
+  podman stop "${CONTAINER_NAME}" >/dev/null 2>&1 || true
+  podman rm -f "${CONTAINER_NAME}" >/dev/null 2>&1 || true
+}
+trap cleanup EXIT
+
+podman rm -f "${CONTAINER_NAME}" >/dev/null 2>&1 || true
+
+HF_ENV=()
+if [[ -n "${HF_TOKEN:-}" ]]; then
+  HF_ENV=(-e "HF_TOKEN=${HF_TOKEN}")
+fi
+
+podman run -d \
+  --name "${CONTAINER_NAME}" \
+  -p "${PORT}:8000" \
+  -v "${CACHE_DIR}:/root/.cache/huggingface:Z" \
+  -e HF_HOME=/root/.cache/huggingface \
+  -e VLLM_CPU_KVCACHE_SPACE=6 \
+  -e VLLM_CPU_NUM_OF_RESERVED_CPU=1 \
+  --security-opt seccomp=unconfined \
+  --cap-add SYS_NICE \
+  --shm-size=4g \
+  "${HF_ENV[@]}" \
+  "${IMAGE}" \
+  "${MODEL}" \
+  --dtype=bfloat16 \
+  --max-model-len 32768 \
+  --api-key "${API_KEY}" >/dev/null
+
+HOST_FQDN="$(hostname -f)"
+export MODEL PORT API_KEY HOST_FQDN
+
+echo "Waiting for service readiness..."
+READY=0
+for i in $(seq 1 240); do
+  if curl -fsS "http://127.0.0.1:${PORT}/v1/models" \
+      -H "Authorization: Bearer ${API_KEY}" >/dev/null 2>&1; then
+    READY=1
+    break
+  fi
+  if (( i % 10 == 0 )); then
+    echo "Still waiting... iteration=${i}"
+  fi
+  sleep 5
+done
+
+if [[ "${READY}" != "1" ]]; then
+  echo "vLLM did not become ready in time"
+  podman logs "${CONTAINER_NAME}" || true
+  exit 1
+fi
+
+echo "Service is ready. Writing registry file..."
+
+python3 - <<'PY'
+import json, os
+jobid = os.environ.get("LSB_JOBID", "manual")
+doc = {
+    "jobid": jobid,
+    "service_name": "qwen-chat",
+    "model": os.environ["MODEL"],
+    "host": os.environ["HOST_FQDN"],
+    "port": int(os.environ["PORT"]),
+    "api_key": os.environ["API_KEY"],
+    "status": "ready",
+}
+path = os.path.expanduser(f"~/lsf_vllm_poc/registry/{jobid}.json")
+with open(path, "w") as f:
+    json.dump(doc, f, indent=2)
+print(f"registry written: {path}")
+print(f"endpoint: http://{doc['host']}:{doc['port']}/v1")
+PY
+
+echo "Streaming container logs..."
+podman logs -f "${CONTAINER_NAME}"