docs/meta: Remove Key Features and add library keywords

Author: thlurte

README.md

@@ -9,16 +9,6 @@ By replacing massive deep learning libraries with highly optimized, compiled C++
 
 ---
 
-## Key Features
-
-- **No PyTorch or Transformers:** Fully decoupled from the heavy standard library pipeline. A simple `pip install` completes in seconds.
-- **Micro Memory Footprint:** Executes multi-vector graphs inside ONNX Runtime, drawing less than 65MB of RAM during inference.
-- **Fast Rust Tokenization:** Uses Hugging Face's raw Rust tokenization backend directly.
-- **Dynamic Punctuation Skiplist:** Dynamically parses `tokenizer.json` at initialization, creating a zero-overhead mask to discard punctuation vectors, matching ColBERT index-saving behaviors.
-- **Standardized Late Interaction:** Exposes native NumPy-based MaxSim calculations.
-
----
-
 ## Installation
 
 Install the library directly via pip:

pyproject.toml

@@ -14,6 +14,18 @@ classifiers = [
     "License :: OSI Approved :: MIT License",
     "Operating System :: OS Independent",
 ]
+keywords = [
+    "colbert",
+    "late-interaction",
+    "onnx",
+    "onnxruntime",
+    "embedding",
+    "search",
+    "information-retrieval",
+    "vector-search",
+    "zero-pytorch",
+    "modernbert"
+]
 dependencies = [
     "onnxruntime>=1.16.0",
     "tokenizers>=0.19.0",