Update requirements.txt

requirements.txt

@@ -1,11 +1,37 @@
-ultralytics
-opencv-python-headless
-pillow
-numpy
-networkx
-pandas
-openai
-huggingface_hub
-easyocr
-scikit-image
-scipy
+# Circuit Component Detector — requirements
+# Install with: pip install -r requirements.txt
+
+# Core app / UI
+gradio>=4.0.0
+
+# Computer vision + detection
+opencv-python>=4.8.0
+ultralytics>=8.0.0          # YOLOv8 (model = YOLO("best.pt"))
+numpy>=1.24.0
+Pillow>=10.0.0
+
+# OCR
+easyocr>=1.7.0
+
+# Graph building (wire-CC adjacency graph, path enumeration)
+networkx>=3.0
+
+# Data handling (CSV comparison feature)
+pandas>=2.0.0
+
+# LLM access — used for BOTH Nemotron calls:
+#   1) get_all_values_with_llm()      -> OCR/label/value reading
+#   2) verify_paths_with_nemotron()   -> NEW: second-opinion path/edge audit
+# Both calls go through OpenRouter's OpenAI-compatible endpoint, so the
+# official `openai` python package is the only client library required.
+# No Anthropic/Claude package needed — this build is Nemotron-only.
+openai>=1.0.0
+
+# Notes:
+# - torch/torchvision are required transitively by ultralytics + easyocr;
+#   pip will resolve them automatically. If you need GPU acceleration for
+#   YOLO, install the CUDA-enabled torch build for your platform BEFORE
+#   running `pip install -r requirements.txt` (see pytorch.org).
+# - easyocr's `gpu=False` is hardcoded in app.py, so OCR will run on CPU
+#   regardless of torch's GPU availability. Change reader = easyocr.Reader(
+#   ['en'], gpu=True) in app.py if you want OCR on GPU too.