diff --git a/.env.example b/.env.example
new file mode 100644
index 0000000..bf58c56
--- /dev/null
+++ b/.env.example
@@ -0,0 +1,13 @@
+# Video processing
+UPLOAD_DIR=./uploads
+OUTPUT_DIR=./output
+MODELS_DIR=./models
+
+# Web UI
+WEB_HOST=0.0.0.0
+WEB_PORT=9000
+SECRET_KEY=change-me
+
+# Detection defaults
+SACK_CONF=0.4
+TRUCK_CONF=0.5
diff --git a/.gitignore b/.gitignore
new file mode 100644
index 0000000..f380588
--- /dev/null
+++ b/.gitignore
@@ -0,0 +1,37 @@
+# Python
+__pycache__/
+*.py[cod]
+*.so
+env/
+venv/
+.venv/
+
+# Environment & state
+.env
+*.db
+
+# Media & outputs (gitignored per global constraints)
+*.mp4
+*.avi
+*.mkv
+*.jpg
+*.jpeg
+*.png
+output/
+uploads/
+
+# TensorRT engines are Jetson build artifacts — rebuildable
+*.engine
+
+# Model weights are reused from karung_counter_semarang and stay local
+models/
+
+# Test artifacts
+.pytest_cache/
+.coverage
+
+# IDE & OS
+.idea/
+.vscode/
+.DS_Store
+Thumbs.db
diff --git a/README.md b/README.md
new file mode 100644
index 0000000..66cb955
--- /dev/null
+++ b/README.md
@@ -0,0 +1,21 @@
+# Feedmill Recounter
+
+AI video analysis tool for counting objects (sacks, boxes) in feedmill videos.
+Built on top of [karung_counter_semarang](https://git.proit.id/andrew/karung-counting-feedmill-semarang).
+
+## Features
+
+- **CLI**: Process videos from the command line with any model + class filter
+- **Web UI**: Upload videos, select models, download annotated output (port 9000)
+- **Multiple Models**: Run multiple model configurations on the same video for comparison
+- **Class Filtering**: Choose which classes to count (sack, box, truck)
+- **Annotated Output**: Download MP4 videos with detection overlays for human review
+
+## Quick Start
+
+```bash
+pip install -e ".[dev]"
+recounter --list-models --models-dir ./models
+recounter-web
+# Open http://localhost:9000
+```
diff --git a/cfg/tracker.yaml b/cfg/tracker.yaml
new file mode 100644
index 0000000..701328a
--- /dev/null
+++ b/cfg/tracker.yaml
@@ -0,0 +1,26 @@
+# Custom FastTrack config tuned for sack counting:
+#   - track_buffer=60: hold lost tracks for 60 frames (~2.4s at 25fps)
+#     to survive worker occlusion
+#   - new_track_thresh=0.3: harder to spawn duplicate IDs
+#   - track_low_thresh=0.05: recover faint detections behind workers
+#   - active_occ_to_lost_thresh=15: tolerate 15 occluded frames
+#   - occ_reappear_window=60: re-find tracks after long occlusion
+#   - enlarge_bbox_occ=1.15: widen search region during occlusion
+
+tracker_type: bytetrack
+track_high_thresh: 0.20
+track_low_thresh: 0.05
+new_track_thresh: 0.30
+track_buffer: 60
+match_thresh: 0.85
+fuse_score: true
+
+# Occlusion handling (FastTrack-specific)
+reset_velocity_offset_occ: 5
+reset_pos_offset_occ: 3
+enlarge_bbox_occ: 1.15
+dampen_motion_occ: 0.4
+active_occ_to_lost_thresh: 15
+occ_cover_thresh: 0.6
+occ_reappear_window: 60
+init_iou_suppress: 0.65
diff --git a/pyproject.toml b/pyproject.toml
new file mode 100644
index 0000000..5c5ffd8
--- /dev/null
+++ b/pyproject.toml
@@ -0,0 +1,27 @@
+[build-system]
+requires = ["setuptools>=68.0"]
+build-backend = "setuptools.build_meta"
+
+[project]
+name = "feedmill-recounter"
+version = "0.1.0"
+description = "AI video analysis tool for counting objects in feedmill videos"
+requires-python = ">=3.10"
+dependencies = [
+    "ultralytics",
+    "opencv-python",
+    "numpy",
+    "shapely",
+    "flask",
+    "python-dotenv",
+]
+
+[project.optional-dependencies]
+dev = ["pytest"]
+
+[project.scripts]
+recounter = "cli:main"
+recounter-web = "app:main"
+
+[tool.pytest.ini_options]
+testpaths = ["tests"]
diff --git a/src/__init__.py b/src/__init__.py
new file mode 100644
index 0000000..e69de29
diff --git a/tests/__init__.py b/tests/__init__.py
new file mode 100644
index 0000000..e69de29