diff --git a/app/.gitattributes b/app/.gitattributes
new file mode 100644
index 0000000..f0bdeab
--- /dev/null
+++ b/app/.gitattributes
@@ -0,0 +1,2 @@
+*.onnx filter=lfs diff=lfs merge=lfs -text
+*.jpg filter=lfs diff=lfs merge=lfs -text
diff --git a/app/Containerfile b/app/Containerfile
new file mode 100644
index 0000000..6f45bb7
--- /dev/null
+++ b/app/Containerfile
@@ -0,0 +1,12 @@
+FROM quay.io/nmasse-redhat/jetpack-multiarch:r36.4.0
+
+ADD *.py *.jpg *.onnx /opt/app
+WORKDIR /opt/app
+RUN <<EOF
+set -Eeu
+pip install -y onnxruntime_gpu
+pip install -r requirements.txt
+EOF
+
+ENTRYPOINT [ "python3", "app.py" ]
+ARG [ ]
diff --git a/app/README.md b/app/README.md
new file mode 100644
index 0000000..c539321
--- /dev/null
+++ b/app/README.md
@@ -0,0 +1,2 @@
+# AI application
+
diff --git a/app/app.py b/app/app.py
new file mode 100644
index 0000000..f31e214
--- /dev/null
+++ b/app/app.py
@@ -0,0 +1,103 @@
+import numpy as np
+import time
+import onnxruntime as ort
+import json
+import os
+import logging
+import ast
+import cv2.dnn
+
+# Environment variables
+MODEL_PATH = os.environ.get("MODEL_PATH", "model.onnx")
+INPUT_IMAGE_PATH = os.environ.get("IMAGE_PATH", "input.jpg")
+
+# Onnxruntime providers
+PROVIDERS = ast.literal_eval(os.environ.get("ONNXRUNTIME_PROVIDERS", '["CUDAExecutionProvider"]'))
+
+CLASSES = {
+  0: "SpeedLimit",
+  1: "DangerAhead"
+}
+colors = np.random.uniform(0, 255, size=(len(CLASSES), 3))
+MIN_CONF_THRESHOLD = float(os.environ.get("MIN_CONF_THRESHOLD", 0.8))
+
+
+def preprocess(original_image):
+    # original_image: np.ndarray = cv2.imread(image_path)
+    [height, width, _] = original_image.shape
+
+    # Prepare a square image for inference
+    length = max((height, width))
+    image = np.zeros((length, length, 3), np.uint8)
+    image[0:height, 0:width] = original_image
+
+    # Calculate scale factor
+    scale = length / 640
+
+    # Preprocess the image and prepare blob for model
+    blob = cv2.dnn.blobFromImage(image, scalefactor=1 / 255, size=(640, 640), swapRB=True)
+    return blob, scale, original_image
+
+def postprocess(response):
+    outputs = np.array([cv2.transpose(response[0])])
+    rows = outputs.shape[1]
+
+    boxes = []
+    scores = []
+    class_ids = []
+
+    # Iterate through output to collect bounding boxes, confidence scores, and class IDs
+    for i in range(rows):
+        classes_scores = outputs[0][i][4:]
+        (minScore, maxScore, minClassLoc, (x, maxClassIndex)) = cv2.minMaxLoc(classes_scores)
+        if maxScore >= 0.25:
+            box = [
+                outputs[0][i][0] - (0.5 * outputs[0][i][2]), outputs[0][i][1] - (0.5 * outputs[0][i][3]),
+                outputs[0][i][2], outputs[0][i][3]]
+            boxes.append(box)
+            scores.append(maxScore)
+            class_ids.append(maxClassIndex)
+
+    detections = []
+    result_boxes = cv2.dnn.NMSBoxes(boxes, scores, 0.25, 0.45, 0.5)
+    # Iterate through NMS results to draw bounding boxes and labels
+    for i in range(len(result_boxes)):
+        index = result_boxes[i]
+        box = boxes[index]
+        if scores[index] > MIN_CONF_THRESHOLD:
+            detection = {
+                'class_id': class_ids[index],
+                'class_name': CLASSES[class_ids[index]],
+                'confidence': f"{scores[index]:.2f}",
+                'box': [f"{c:.2f}" for c in box]}
+            detections.append(detection)
+    return detections
+
+if __name__ == "__main__":
+    # Logger
+    logging.basicConfig(
+        format='%(asctime)s %(levelname)-8s %(message)s',
+        level=logging.INFO,
+        datefmt='%Y-%m-%d %H:%M:%S'
+    )
+    logger = logging.getLogger(__name__)
+    ort_sess = ort.InferenceSession(MODEL_PATH, providers=PROVIDERS)
+    nparr = np.fromfile(INPUT_IMAGE_PATH, np.uint8)
+    nparr = cv2.imdecode(nparr, cv2.IMREAD_COLOR)
+    preprocessed, scale, original_image = preprocess(nparr)
+    outputs = ort_sess.run(None, {'images': preprocessed})
+    detections = postprocess(outputs[0])
+    logger.info(f"Processed image {INPUT_IMAGE_PATH}")
+    logger.info(json.dumps(detections))
+    logger.info(f"Now starting mass inference...")
+    start = time.time()
+    count = 0
+    while True:
+        now = time.time()
+        if now - start > 1:
+            logger.info(f"Performed {count} inferences in {now-start:.2f}s")
+            start = time.time()
+            count = 0
+        outputs = ort_sess.run(None, {'images': preprocessed})
+        count+=1
+
diff --git a/app/input.jpg b/app/input.jpg
new file mode 100644
index 0000000..59eb32b
--- /dev/null
+++ b/app/input.jpg
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:737bac004419fa919c8ca2f17b73e7e51df11753d5acf866f5e8d92c2d5858bf
+size 48350
diff --git a/app/model.onnx b/app/model.onnx
new file mode 100644
index 0000000..4a89cd2
--- /dev/null
+++ b/app/model.onnx
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b324341eac87dd48c4d301afdc1d3aa7d58a8df86ec789b86d02f52bdf3ad192
+size 12238986
diff --git a/app/requirements.txt b/app/requirements.txt
new file mode 100644
index 0000000..f9a892d
--- /dev/null
+++ b/app/requirements.txt
@@ -0,0 +1,3 @@
+opencv-python-headless
+numpy
+onnxruntime