diff --git a/examples/quickstart/README.md b/examples/quickstart/README.md
new file mode 100644
index 00000000..4e2853c5
--- /dev/null
+++ b/examples/quickstart/README.md
@@ -0,0 +1,47 @@
+# Semantic Router Quickstart
+
+> ⚠️ This is an initial skeleton for the 10-minute quickstart workflow. Content will be expanded in follow-up tasks.
+
+## Goal
+Provide a single command path (`make quickstart`) that prepares the router, runs a small evaluation, and surfaces a concise report so that new users can validate the system within 10 minutes.
+
+## Structure
+- `quickstart.sh` – orchestrates dependency checks, model downloads, and service startup.
+- `quick-eval.sh` – executes a minimal benchmark run and captures results.
+- `config-quickstart.yaml` – opinionated defaults for running the router locally.
+- `sample-data/` – trimmed datasets used for fast evaluation.
+- `templates/` – report and config templates shared by quickstart scripts.
+
+## Next Steps
+1. Teach the benchmark loader to honor `QUICKSTART_SAMPLE_ROOT` so local JSONL slices are used offline.
+2. Add a Makefile target (`quickstart`) that chains router bootstrap and quick evaluation.
+3. Create CI smoke tests that run the 10-minute flow with the trimmed datasets.
+
+## Quick Evaluation
+Run the standalone evaluator once the router is healthy. A typical flow looks like:
+
+```bash
+./examples/quickstart/quickstart.sh &   # starts router (Ctrl+C to stop)
+./examples/quickstart/quick-eval.sh --dataset mmlu --samples 5 --mode router
+```
+
+The evaluation script will place raw artifacts under `examples/quickstart/results/<timestamp>/raw` and derive:
+- `quickstart-summary.csv` – compact metrics table for spreadsheets or dashboards.
+- `quickstart-report.md` – Markdown summary suitable for PRs or runbooks.
+
+Key flags:
+- `--mode router|vllm|both` to toggle which side runs.
+- `--samples` to tune runtime vs. statistical confidence.
+- `--output-dir` for custom destinations (defaults to timestamped folder).
+- All settings also respect `QUICKSTART_*` environment overrides.
+
+## Local Sample Data
+The `sample-data/` directory now includes trimmed JSONL slices for quick runs:
+- `mmlu-sample.jsonl` – 10 multi-category academic questions.
+- `arc-sample.jsonl` – 10 middle-school science questions with ARC-style options.
+
+Each record follows the same schema that the benchmark loader expects (`question_id`, `category`, `question`, `options`, `answer`, optional `cot_content`). Sizes stay under 10 KB per file so the quickstart remains lightweight.
+
+**Integration hook**: upcoming work will extend `bench/vllm_semantic_router_bench` to read from these JSONL files whenever `QUICKSTART_SAMPLE_ROOT` is set (falling back to Hugging Face datasets otherwise). Keep the files committed and deterministic so that the automated 10-minute flow can depend on them once the loader change lands.
+
+
diff --git a/examples/quickstart/config-quickstart.yaml b/examples/quickstart/config-quickstart.yaml
new file mode 100644
index 00000000..9d7552d7
--- /dev/null
+++ b/examples/quickstart/config-quickstart.yaml
@@ -0,0 +1,90 @@
+# Quickstart configuration tuned for a single-node developer setup.
+# Keeps routing options minimal while remaining compatible with the default assets
+# shipped by `make download-models`.
+
+bert_model:
+  model_id: sentence-transformers/all-MiniLM-L12-v2
+  threshold: 0.6
+  use_cpu: true
+
+semantic_cache:
+  enabled: false
+  backend_type: "memory"
+
+prompt_guard:
+  enabled: false
+  use_modernbert: true
+  model_id: "models/jailbreak_classifier_modernbert-base_model"
+  threshold: 0.7
+  use_cpu: true
+  jailbreak_mapping_path: "models/jailbreak_classifier_modernbert-base_model/jailbreak_type_mapping.json"
+
+classifier:
+  category_model:
+    model_id: "models/category_classifier_modernbert-base_model"
+    threshold: 0.6
+    use_cpu: true
+    use_modernbert: true
+    category_mapping_path: "models/category_classifier_modernbert-base_model/category_mapping.json"
+  pii_model:
+    model_id: "models/pii_classifier_modernbert-base_presidio_token_model"
+    threshold: 0.7
+    use_cpu: true
+    pii_mapping_path: "models/pii_classifier_modernbert-base_presidio_token_model/pii_type_mapping.json"
+
+vllm_endpoints:
+  - name: "local-vllm"
+    address: "127.0.0.1"
+    port: 8000
+    models:
+      - "openai/gpt-oss-20b"
+    weight: 1
+
+model_config:
+  "openai/gpt-oss-20b":
+    preferred_endpoints: ["local-vllm"]
+    reasoning_family: "gpt-oss"
+    pii_policy:
+      allow_by_default: true
+
+categories:
+  - name: general
+    system_prompt: "You are a helpful and knowledgeable assistant. Provide concise, accurate answers."
+    model_scores:
+      - model: openai/gpt-oss-20b
+        score: 0.7
+        use_reasoning: false
+
+  - name: reasoning
+    system_prompt: "You explain your reasoning with clear numbered steps before giving a final answer."
+    model_scores:
+      - model: openai/gpt-oss-20b
+        score: 0.6
+        use_reasoning: true
+
+  - name: safety
+    system_prompt: "You prioritize safe completions and refuse harmful requests."
+    model_scores:
+      - model: openai/gpt-oss-20b
+        score: 0.5
+        use_reasoning: false
+
+default_model: openai/gpt-oss-20b
+
+reasoning_families:
+  gpt-oss:
+    type: "chat_template_kwargs"
+    parameter: "thinking"
+
+api:
+  batch_classification:
+    metrics:
+      enabled: false
+
+# Tool auto-selection is available but disabled for quickstart.
+tools:
+  enabled: false
+  top_k: 3
+  similarity_threshold: 0.2
+  tools_db_path: "config/tools_db.json"
+  fallback_to_empty: true
diff --git a/examples/quickstart/quick-eval.sh b/examples/quickstart/quick-eval.sh
new file mode 100755
index 00000000..274e6d92
--- /dev/null
+++ b/examples/quickstart/quick-eval.sh
@@ -0,0 +1,292 @@
+#!/usr/bin/env bash
+set -euo pipefail
+
+SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
+ROOT_DIR="$(cd "$SCRIPT_DIR/../.." && pwd)"
+RESULTS_ROOT="${SCRIPT_DIR}/results"
+DEFAULT_DATASET="mmlu"
+DEFAULT_SAMPLES=${QUICKSTART_SAMPLES:-5}
+DEFAULT_MODE="${QUICKSTART_MODE:-router}"
+DEFAULT_ROUTER_MODELS="${QUICKSTART_ROUTER_MODELS:-auto}"
+DEFAULT_VLLM_MODELS="${QUICKSTART_VLLM_MODELS:-openai/gpt-oss-20b}"
+DEFAULT_ROUTER_ENDPOINT="${ROUTER_ENDPOINT:-http://127.0.0.1:8801/v1}"
+DEFAULT_VLLM_ENDPOINT="${VLLM_ENDPOINT:-http://127.0.0.1:8000/v1}"
+DEFAULT_SEED=${QUICKSTART_SEED:-42}
+REQUIRED_COMMANDS=()
+REQUIRED_PY_MODULES=(numpy pandas datasets openai)
+
+PYTHON_BIN="${PYTHON_BIN:-}"
+if [[ -z "$PYTHON_BIN" ]]; then
+  if [[ -n "${VIRTUAL_ENV:-}" && -x "${VIRTUAL_ENV}/bin/python3" ]]; then
+    PYTHON_BIN="${VIRTUAL_ENV}/bin/python3"
+  elif [[ -n "${VIRTUAL_ENV:-}" && -x "${VIRTUAL_ENV}/bin/python" ]]; then
+    PYTHON_BIN="${VIRTUAL_ENV}/bin/python"
+  elif command -v python3 >/dev/null 2>&1; then
+    PYTHON_BIN="$(command -v python3)"
+  elif command -v python >/dev/null 2>&1; then
+    PYTHON_BIN="$(command -v python)"
+  else
+    echo '[ERROR] No python interpreter found in PATH or VIRTUAL_ENV' >&2
+    exit 1
+  fi
+fi
+
+usage() {
+  cat <<'USAGE'
+Usage: quick-eval.sh [OPTIONS]
+
+Options:
+  --dataset NAME          Dataset identifier (default: mmlu)
+  --samples N             Samples per category (default: 5 or $QUICKSTART_SAMPLES)
+  --mode MODE             router|vllm|both (default: router or $QUICKSTART_MODE)
+  --router-endpoint URL   Router endpoint (default: env ROUTER_ENDPOINT or http://127.0.0.1:8801/v1)
+  --vllm-endpoint URL     vLLM endpoint (default: env VLLM_ENDPOINT or http://127.0.0.1:8000/v1)
+  --router-models LIST    Space-separated router models (default: auto)
+  --vllm-models LIST      Space-separated vLLM models (default: openai/gpt-oss-20b)
+  --output-dir DIR        Directory to store run artifacts (default: examples/quickstart/results/<timestamp>)
+  --seed N                Random seed (default: 42 or $QUICKSTART_SEED)
+  --help                  Show this help message
+
+Environment overrides:
+  QUICKSTART_SAMPLES, QUICKSTART_MODE, QUICKSTART_ROUTER_MODELS,
+  QUICKSTART_VLLM_MODELS, QUICKSTART_SEED, ROUTER_ENDPOINT, VLLM_ENDPOINT.
+
+The script launches the benchmark module with quickstart-friendly defaults and
+emits both CSV and Markdown summaries for the generated results.
+USAGE
+}
+
+log() {
+  local level="$1"
+  shift
+  printf '[%s] %s\n' "$level" "$*"
+}
+
+die() {
+  log "ERROR" "$*"
+  exit 1
+}
+
+parse_args() {
+  DATASET="$DEFAULT_DATASET"
+  SAMPLES="$DEFAULT_SAMPLES"
+  MODE="$DEFAULT_MODE"
+  ROUTER_ENDPOINT="$DEFAULT_ROUTER_ENDPOINT"
+  VLLM_ENDPOINT="$DEFAULT_VLLM_ENDPOINT"
+  ROUTER_MODELS=("$DEFAULT_ROUTER_MODELS")
+  VLLM_MODELS=("$DEFAULT_VLLM_MODELS")
+  OUTPUT_DIR=""
+  SEED="$DEFAULT_SEED"
+
+  while [[ $# -gt 0 ]]; do
+    case "$1" in
+      --dataset)
+        DATASET="$2"
+        shift 2
+        ;;
+      --samples)
+        SAMPLES="$2"
+        shift 2
+        ;;
+      --mode)
+        MODE="$2"
+        shift 2
+        ;;
+      --router-endpoint)
+        ROUTER_ENDPOINT="$2"
+        shift 2
+        ;;
+      --vllm-endpoint)
+        VLLM_ENDPOINT="$2"
+        shift 2
+        ;;
+      --router-models)
+        shift
+        ROUTER_MODELS=()
+        while [[ $# -gt 0 ]] && [[ $1 != --* ]]; do
+          ROUTER_MODELS+=("$1")
+          shift
+        done
+        ;;
+      --vllm-models)
+        shift
+        VLLM_MODELS=()
+        while [[ $# -gt 0 ]] && [[ $1 != --* ]]; do
+          VLLM_MODELS+=("$1")
+          shift
+        done
+        ;;
+      --output-dir)
+        OUTPUT_DIR="$2"
+        shift 2
+        ;;
+      --seed)
+        SEED="$2"
+        shift 2
+        ;;
+      --help|-h)
+        usage
+        exit 0
+        ;;
+      *)
+        die "Unknown option: $1"
+        ;;
+    esac
+  done
+
+  case "$MODE" in
+    router|vllm|both) ;;
+    *) die "Invalid --mode '$MODE' (expected router|vllm|both)" ;;
+  esac
+
+  if [[ -z "$OUTPUT_DIR" ]]; then
+    local timestamp
+    timestamp="$(date +%Y%m%d_%H%M%S)"
+    OUTPUT_DIR="${RESULTS_ROOT}/${timestamp}"
+  fi
+}
+
+require_commands() {
+  local missing=()
+  for cmd in "${REQUIRED_COMMANDS[@]}"; do
+    if ! command -v "$cmd" >/dev/null 2>&1; then
+      missing+=("$cmd")
+    fi
+  done
+  if [[ ${#missing[@]} -gt 0 ]]; then
+    die "Missing required commands: ${missing[*]}"
+  fi
+}
+
+check_python_modules() {
+  if ! "$PYTHON_BIN" - "${REQUIRED_PY_MODULES[@]}" <<'PY'; then
+import importlib
+import sys
+missing = []
+for name in sys.argv[1:]:
+    try:
+        importlib.import_module(name)
+    except Exception:
+        missing.append(name)
+if missing:
+    raise SystemExit("Missing Python modules: " + ", ".join(missing))
+PY
+    die "Missing Python modules. Run 'pip install -r bench/requirements.txt'"
+  fi
+}
+
+prepare_dirs() {
+  mkdir -p "$RESULTS_ROOT"
+  mkdir -p "$OUTPUT_DIR"
+}
+
+run_benchmark() {
+  local raw_dir="${OUTPUT_DIR}/raw"
+  mkdir -p "$raw_dir"
+
+  local cmd=("$PYTHON_BIN" -m vllm_semantic_router_bench.router_reason_bench_multi_dataset
+    --dataset "$DATASET"
+    --samples-per-category "$SAMPLES"
+    --output-dir "$raw_dir"
+    --seed "$SEED"
+    --router-endpoint "$ROUTER_ENDPOINT"
+  )
+
+  if [[ "$MODE" == "router" || "$MODE" == "both" ]]; then
+    cmd+=(--run-router --router-models "${ROUTER_MODELS[@]}")
+  fi
+
+  if [[ "$MODE" == "vllm" || "$MODE" == "both" ]]; then
+    cmd+=(--run-vllm --vllm-endpoint "$VLLM_ENDPOINT" --vllm-models "${VLLM_MODELS[@]}" --vllm-exec-modes NR)
+  fi
+
+  log "INFO" "Running benchmark: ${cmd[*]}"
+  (
+    cd "$ROOT_DIR"
+    PYTHONPATH="$ROOT_DIR/bench${PYTHONPATH:+:$PYTHONPATH}" \
+    ROUTER_ENDPOINT="$ROUTER_ENDPOINT" \
+    VLLM_ENDPOINT="$VLLM_ENDPOINT" \
+      "${cmd[@]}"
+  ) || die "Benchmark run failed"
+
+  RAW_DIR="$raw_dir"
+}
+
+collect_summaries() {
+  mapfile -t SUMMARY_FILES < <(find "$RAW_DIR" -type f -name summary.json -print | sort)
+  if [[ ${#SUMMARY_FILES[@]} -eq 0 ]]; then
+    die "No summary.json files produced under $RAW_DIR"
+  fi
+
+  SUMMARY_CSV="$OUTPUT_DIR/quickstart-summary.csv"
+  REPORT_MD="$OUTPUT_DIR/quickstart-report.md"
+
+  "$PYTHON_BIN" - "$SUMMARY_CSV" "$REPORT_MD" "${SUMMARY_FILES[@]}" <<'PY'
+import json
+import sys
+from datetime import datetime
+from pathlib import Path
+
+summary_csv = Path(sys.argv[1])
+report_md = Path(sys.argv[2])
+summary_paths = [Path(p) for p in sys.argv[3:]]
+
+rows = []
+for path in summary_paths:
+    data = json.loads(path.read_text())
+    rows.append({
+        "dataset": data.get("dataset"),
+        "model": data.get("model"),
+        "overall_accuracy": data.get("overall_accuracy"),
+        "avg_response_time": data.get("avg_response_time"),
+        "avg_total_tokens": data.get("avg_total_tokens"),
+        "total_questions": data.get("total_questions"),
+        "successful_queries": data.get("successful_queries"),
+        "failed_queries": data.get("failed_queries"),
+        "source": str(path.relative_to(summary_csv.parent))
+    })
+
+rows.sort(key=lambda r: (r["dataset"] or "", r["model"] or ""))
+
+csv_lines = ["dataset,model,overall_accuracy,avg_response_time,avg_total_tokens,total_questions,successful_queries,failed_queries,source"]
+for row in rows:
+    csv_lines.append(
+        f"{row['dataset']},{row['model']},{row['overall_accuracy']},{row['avg_response_time']},{row['avg_total_tokens']},{row['total_questions']},{row['successful_queries']},{row['failed_queries']},{row['source']}"
+    )
+summary_csv.write_text("\n".join(csv_lines) + "\n")
+
+lines = ["# Quickstart Evaluation Report", "", f"Generated: {datetime.utcnow().isoformat()}Z", ""]
+lines.append("| Dataset | Model | Accuracy | Avg Latency (s) | Avg Tokens | Samples |")
+lines.append("| --- | --- | --- | --- | --- | --- |")
+for row in rows:
+    accuracy = f"{row['overall_accuracy']:.3f}" if isinstance(row['overall_accuracy'], (int, float)) else "N/A"
+    latency = f"{row['avg_response_time']:.2f}" if isinstance(row['avg_response_time'], (int, float)) else "N/A"
+    tokens = f"{row['avg_total_tokens']:.1f}" if isinstance(row['avg_total_tokens'], (int, float)) else "N/A"
+    total = row['total_questions'] if row['total_questions'] is not None else 0
+    success = row['successful_queries'] if row['successful_queries'] is not None else 0
+    lines.append(f"| {row['dataset']} | {row['model']} | {accuracy} | {latency} | {tokens} | {success}/{total} |")
+
+lines.append("")
+lines.append("## Source Artifacts")
+lines.append("")
+for row in rows:
+    lines.append(f"- `{row['source']}`")
+
+report_md.write_text("\n".join(lines) + "\n")
+PY
+
+  log "INFO" "Summary CSV: $SUMMARY_CSV"
+  log "INFO" "Report Markdown: $REPORT_MD"
+}
+
+main() {
+  parse_args "$@"
+  require_commands
+  log "INFO" "Using python interpreter: ${PYTHON_BIN}"
+  check_python_modules
+  prepare_dirs
+  run_benchmark
+  collect_summaries
+}
+
+main "$@"
diff --git a/examples/quickstart/quickstart.sh b/examples/quickstart/quickstart.sh
new file mode 100755
index 00000000..dec978ce
--- /dev/null
+++ b/examples/quickstart/quickstart.sh
@@ -0,0 +1,192 @@
+#!/usr/bin/env bash
+set -euo pipefail
+
+SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
+ROOT_DIR="$(cd "$SCRIPT_DIR/../.." && pwd)"
+CONFIG_FILE="${SCRIPT_DIR}/config-quickstart.yaml"
+ROUTER_BIN="${ROOT_DIR}/bin/router"
+ROUTER_LOG_DIR="${SCRIPT_DIR}/logs"
+ROUTER_LOG_FILE="${ROUTER_LOG_DIR}/router.log"
+HEALTH_URL="${QUICKSTART_HEALTH_URL:-http://127.0.0.1:8801/health}"
+REQUIRED_COMMANDS=(make go cargo rustc python3 curl)
+SKIP_DOWNLOAD=false
+SKIP_BUILD=false
+SKIP_START=false
+ROUTER_HEALTH_TIMEOUT=${QUICKSTART_HEALTH_TIMEOUT:-60}
+
+usage() {
+  cat <<'USAGE'
+Usage: quickstart.sh [OPTIONS]
+
+Options:
+  --skip-download   Do not run `make download-models` (expects assets present)
+  --skip-build      Do not run `make build`
+  --skip-start      Skip starting the router (useful for smoke builds)
+  -h, --help        Show this help message
+
+Environment overrides:
+  QUICKSTART_HEALTH_URL       Health probe URL (default http://127.0.0.1:8801/health)
+  QUICKSTART_HEALTH_TIMEOUT   Seconds to wait for router health (default 60)
+USAGE
+}
+
+log() {
+  local level="$1"
+  shift
+  printf '[%s] %s\n' "$level" "$*"
+}
+
+die() {
+  log "ERROR" "$*"
+  exit 1
+}
+
+parse_args() {
+  while [[ $# -gt 0 ]]; do
+    case "$1" in
+      --skip-download)
+        SKIP_DOWNLOAD=true
+        shift
+        ;;
+      --skip-build)
+        SKIP_BUILD=true
+        shift
+        ;;
+      --skip-start)
+        SKIP_START=true
+        shift
+        ;;
+      -h|--help)
+        usage
+        exit 0
+        ;;
+      *)
+        die "Unknown option: $1"
+        ;;
+    esac
+  done
+}
+
+require_commands() {
+  local missing=()
+  for cmd in "${REQUIRED_COMMANDS[@]}"; do
+    if ! command -v "$cmd" >/dev/null 2>&1; then
+      missing+=("$cmd")
+    fi
+  done
+
+  if [[ ${#missing[@]} -gt 0 ]]; then
+    log "ERROR" "Missing required commands: ${missing[*]}"
+    log "ERROR" "Install the missing tooling and re-run quickstart."
+    exit 1
+  fi
+}
+
+ensure_config_ready() {
+  if [[ ! -f "$CONFIG_FILE" ]]; then
+    die "Expected config at $CONFIG_FILE"
+  fi
+
+  if grep -q "TODO" "$CONFIG_FILE"; then
+    log "WARN" "Config file contains TODO markers; router bootstrap may fail until populated."
+  fi
+}
+
+run_make_target() {
+  local target="$1"
+  shift || true
+  log "INFO" "Running make $target"
+  (cd "$ROOT_DIR" && make "$target" "$@") || die "make $target failed"
+}
+
+build_assets() {
+  if [[ "$SKIP_DOWNLOAD" != true ]]; then
+    run_make_target download-models
+  else
+    log "INFO" "Skipping model download as requested"
+  fi
+
+  if [[ "$SKIP_BUILD" != true ]]; then
+    run_make_target build
+  else
+    log "INFO" "Skipping build as requested"
+  fi
+}
+
+wait_for_health() {
+  local elapsed=0
+  while (( elapsed < ROUTER_HEALTH_TIMEOUT )); do
+    if curl -fsS --max-time 2 "$HEALTH_URL" >/dev/null 2>&1; then
+      log "INFO" "Router is healthy at $HEALTH_URL"
+      return 0
+    fi
+    sleep 1
+    ((elapsed++))
+  done
+
+  log "ERROR" "Router failed health check within ${ROUTER_HEALTH_TIMEOUT}s"
+  if [[ -f "$ROUTER_LOG_FILE" ]]; then
+    log "INFO" "Tail of router log:"
+    tail -n 40 "$ROUTER_LOG_FILE"
+  fi
+  return 1
+}
+
+cleanup_router() {
+  if [[ -n "${ROUTER_PID:-}" ]] && kill -0 "$ROUTER_PID" >/dev/null 2>&1; then
+    log "INFO" "Stopping router (PID $ROUTER_PID)"
+    kill "$ROUTER_PID" >/dev/null 2>&1 || true
+    wait "$ROUTER_PID" 2>/dev/null || true
+  fi
+}
+
+start_router() {
+  ensure_config_ready
+  mkdir -p "$ROUTER_LOG_DIR"
+
+  if [[ ! -x "$ROUTER_BIN" ]]; then
+    die "Router binary not found at $ROUTER_BIN; run with --skip-start to build only"
+  fi
+
+  log "INFO" "Launching router with $CONFIG_FILE"
+  local ld_path="${ROOT_DIR}/candle-binding/target/release"
+  (
+    cd "$ROOT_DIR" || exit 1
+    export LD_LIBRARY_PATH="$ld_path${LD_LIBRARY_PATH:+:$LD_LIBRARY_PATH}"
+    export DYLD_LIBRARY_PATH="$ld_path${DYLD_LIBRARY_PATH:+:$DYLD_LIBRARY_PATH}"
+    exec "$ROUTER_BIN" -config="$CONFIG_FILE"
+  ) >"$ROUTER_LOG_FILE" 2>&1 &
+
+  ROUTER_PID=$!
+  if [[ -z "${ROUTER_PID}" ]]; then
+    die "Failed to spawn router"
+  fi
+  log "INFO" "Router PID: $ROUTER_PID (logs: $ROUTER_LOG_FILE)"
+
+  if ! wait_for_health; then
+    cleanup_router
+    die "Router did not become healthy"
+  fi
+
+  log "INFO" "Router started successfully. Press Ctrl+C to stop."
+  wait "$ROUTER_PID"
+}
+
+main() {
+  parse_args "$@"
+  require_commands
+  build_assets
+
+  if [[ "$SKIP_START" == true ]]; then
+    log "INFO" "Skipping router startup; quickstart build complete"
+    exit 0
+  fi
+
+  trap 'cleanup_router; exit 130' INT
+  trap 'cleanup_router; exit 143' TERM
+  trap 'cleanup_router' EXIT
+
+  start_router
+}
+
+main "$@"
diff --git a/examples/quickstart/sample-data/README.md b/examples/quickstart/sample-data/README.md
new file mode 100644
index 00000000..94963d2b
--- /dev/null
+++ b/examples/quickstart/sample-data/README.md
@@ -0,0 +1,24 @@
+# Sample Data
+
+Trimmed datasets for the quickstart flow live here. They mirror the schema consumed by
+`bench/vllm_semantic_router_bench` so we can swap them in for the heavy upstream corpora.
+
+## Available slices
+- `mmlu-sample.jsonl` – 10 representative questions spanning multiple disciplines.
+- `arc-sample.jsonl` – 10 ARC-style science questions with multiple-choice options.
+
+Each line is a JSON object with the following fields:
+- `question_id`: stable identifier used in reports.
+- `category`: benchmark category or sub-domain.
+- `question`: user prompt text.
+- `options`: either a list (MMLU) or mapping (ARC) of answer choices.
+- `answer`: canonical correct answer (letter).
+- `cot_content` *(optional)*: short reasoning snippet for CoT prompts.
+
+Files intentionally stay under 20 KB so they can ship with the repository.
+
+## Integration plan
+`quick-eval.sh` will soon surface a `QUICKSTART_SAMPLE_ROOT` override. When present,
+the benchmark runner will load these JSONL files instead of fetching full datasets
+from Hugging Face. Until that wire-up lands, the JSONL fixtures exist here to unblock
+concurrent work (documentation, UI previews, and benchmarking harness updates).
diff --git a/examples/quickstart/sample-data/arc-sample.jsonl b/examples/quickstart/sample-data/arc-sample.jsonl
new file mode 100644
index 00000000..cd94c34b
--- /dev/null
+++ b/examples/quickstart/sample-data/arc-sample.jsonl
@@ -0,0 +1,10 @@
+{"question_id":"dev-arc-001","category":"physical science","question":"Which phase change occurs when a solid turns directly into a gas?","options":{"A":"Condensation","B":"Sublimation","C":"Deposition","D":"Evaporation"},"answer":"B","cot_content":"Solid to gas without liquid is sublimation."}
+{"question_id":"dev-arc-002","category":"earth science","question":"What causes day and night on Earth?","options":{"A":"Earth's revolution around the Sun","B":"Earth's rotation on its axis","C":"Moon's gravitational pull","D":"Seasonal tilt"},"answer":"B","cot_content":"Earth spins on its axis every 24 hours creating day/night."}
+{"question_id":"dev-arc-003","category":"life science","question":"Which structure transports water in vascular plants?","options":{"A":"Phloem","B":"Xylem","C":"Stomata","D":"Chloroplast"},"answer":"B","cot_content":"Xylem carries water and minerals upward."}
+{"question_id":"dev-arc-004","category":"physical science","question":"An object with a mass of 5 kg is pushed with a force of 20 N. What is its acceleration?","options":{"A":"1 m/s^2","B":"2 m/s^2","C":"4 m/s^2","D":"5 m/s^2"},"answer":"C","cot_content":"F=ma => a = 20/5 = 4 m/s²."}
+{"question_id":"dev-arc-005","category":"life science","question":"Which organ system is responsible for transporting nutrients and oxygen throughout the body?","options":{"A":"Digestive","B":"Circulatory","C":"Respiratory","D":"Nervous"},"answer":"B","cot_content":"Circulatory system moves blood carrying nutrients and oxygen."}
+{"question_id":"dev-arc-006","category":"earth science","question":"What is the primary cause of the changing seasons?","options":{"A":"Earth's distance from the Sun","B":"Earth's axial tilt","C":"Solar flares","D":"Ocean currents"},"answer":"B","cot_content":"Tilt causes different hemispheres to receive varying sunlight."}
+{"question_id":"dev-arc-007","category":"energy","question":"Which energy transformation occurs in a flashlight when it is switched on?","options":{"A":"Electrical to light","B":"Mechanical to electrical","C":"Chemical to mechanical","D":"Thermal to electrical"},"answer":"A","cot_content":"Battery provides electrical energy converted to light."}
+{"question_id":"dev-arc-008","category":"physical science","question":"What force keeps the planets in orbit around the Sun?","options":{"A":"Electromagnetic","B":"Frictional","C":"Gravitational","D":"Nuclear"},"answer":"C","cot_content":"Gravity provides the centripetal force for orbits."}
+{"question_id":"dev-arc-009","category":"life science","question":"Which part of the cell contains genetic material?","options":{"A":"Mitochondria","B":"Cell membrane","C":"Nucleus","D":"Ribosome"},"answer":"C","cot_content":"DNA is stored inside the nucleus."}
+{"question_id":"dev-arc-010","category":"earth science","question":"What instrument measures atmospheric pressure?","options":{"A":"Thermometer","B":"Barometer","C":"Hygrometer","D":"Anemometer"},"answer":"B","cot_content":"Barometer readings track air pressure."}
diff --git a/examples/quickstart/sample-data/mmlu-sample.jsonl b/examples/quickstart/sample-data/mmlu-sample.jsonl
new file mode 100644
index 00000000..408cb95f
--- /dev/null
+++ b/examples/quickstart/sample-data/mmlu-sample.jsonl
@@ -0,0 +1,10 @@
+{"question_id":"dev-mmlu-001","category":"business","question":"A firm increases production from 100 to 120 units while total costs rise from $200 to $230. What is the marginal cost per unit?","options":["$1.50","$2.50","$3.00","$15.00"],"answer":"B","cot_content":"Marginal cost = change in total cost / change in quantity = 30 / 20 = 1.5"}
+{"question_id":"dev-mmlu-002","category":"biology","question":"Which organelle is primarily responsible for ATP production in eukaryotic cells?","options":["Nucleus","Endoplasmic reticulum","Mitochondrion","Golgi apparatus"],"answer":"C","cot_content":"ATP synthesis occurs in mitochondria via oxidative phosphorylation."}
+{"question_id":"dev-mmlu-003","category":"computer science","question":"In Big-O notation, the time complexity of binary search on a sorted array is?","options":["O(n)","O(log n)","O(n log n)","O(1)"],"answer":"B","cot_content":"Binary search halves the search space each step, giving logarithmic complexity."}
+{"question_id":"dev-mmlu-004","category":"law","question":"Which principle requires courts to follow precedents set by higher courts?","options":["Mens rea","Stare decisis","Ultra vires","Res ipsa loquitur"],"answer":"B","cot_content":"Stare decisis compels adherence to higher court decisions."}
+{"question_id":"dev-mmlu-005","category":"mathematics","question":"Evaluate the derivative of f(x)=3x^2+4x-5.","options":["3x^2","6x+4","3x+4","6x"],"answer":"B","cot_content":"Derivative is 6x + 4."}
+{"question_id":"dev-mmlu-006","category":"psychology","question":"Which psychologist is associated with the hierarchy of needs?","options":["Carl Rogers","Sigmund Freud","Abraham Maslow","B.F. Skinner"],"answer":"C","cot_content":"Maslow proposed the hierarchy of needs."}
+{"question_id":"dev-mmlu-007","category":"economics","question":"If the price elasticity of demand is -2, a 5% price increase results in what approximate change in quantity demanded?","options":["Increase 2.5%","Decrease 10%","Decrease 2.5%","Increase 10%"],"answer":"B","cot_content":"Elasticity * price change = -2 * 5% = -10%."}
+{"question_id":"dev-mmlu-008","category":"physics","question":"What is the SI unit of electric resistance?","options":["Tesla","Ohm","Farad","Weber"],"answer":"B","cot_content":"Resistance is measured in ohms."}
+{"question_id":"dev-mmlu-009","category":"medicine","question":"Which vitamin deficiency causes scurvy?","options":["Vitamin A","Vitamin B12","Vitamin C","Vitamin D"],"answer":"C","cot_content":"Vitamin C deficiency leads to scurvy."}
+{"question_id":"dev-mmlu-010","category":"history","question":"The Marshall Plan aimed to rebuild which region after WWII?","options":["South America","Eastern Asia","Western Europe","Middle East"],"answer":"C","cot_content":"The Marshall Plan targeted Western European recovery."}
diff --git a/examples/quickstart/templates/README.md b/examples/quickstart/templates/README.md
new file mode 100644
index 00000000..a6dcbbb7
--- /dev/null
+++ b/examples/quickstart/templates/README.md
@@ -0,0 +1,7 @@
+# Quickstart Templates
+
+This folder will include Markdown or YAML templates consumed by quickstart scripts, such as:
+- `report-template.md` for summarizing benchmark output.
+- `config-template.yaml` containing reusable sections shared across quickstart configs.
+
+Add concrete templates once reporting format is finalized.