Add links to paper/code and a usage example

README.md

@@ -1,16 +1,20 @@
 ---
 license: apache-2.0
+size_categories:
+- n<1K
 task_categories:
 - video-text-to-text
 tags:
 - egocentric
 - egoproactive
 - video
-size_categories:
-- n<1K
 ---
+
 # EgoProactive synthetic proactive annotations
 
+**Paper:** [Ambient @ EgoProactive 2026 : Proactive Egocentric Assistance with Visually Grounded Supervision](https://huggingface.co/papers/2609.07099)  
+**Code:** [ambient-intelligence-hq/egoproactive-verbalizer](https://github.com/ambient-intelligence-hq/egoproactive-verbalizer)
+
 Everything produced by the annotation and synthesis pipelines for the [AI Wearables Challenge 2026](https://wearable-ai-workshop.github.io/) EgoProactive
 
 Dense timestamped proactive walkthroughs generated with the **ambient agent**
@@ -31,4 +35,13 @@ format: `answers[j]` = `$interrupt$<utterance>` | `$silent$` aligned to `video_i
 
 Measured effect (Qwen3.5-4B verbalizer, LoRA r32): egoconv-dense **+0.021 G-mean cross-domain**
 (HoloAssist-140: 0.557 -> 0.577, interrupt-F1 +0.064) but **-0.042 in-domain** (egoproactive-140)
-— i.e. it trades in-domain fit for out-of-distribution robustness.
+— i.e. it trades in-domain fit for out-of-distribution robustness.
+
+## Dataset Usage
+
+```bash
+huggingface-cli download ambient-intelligence-labs/egoproactive-synth-annotations --repo-type dataset \
+  --include "val700_train.jsonl" "egoconv_dense_train.jsonl" --local-dir data/
+```
+
+For full reproduction steps (frame extraction, training, evaluation) see the [GitHub repository](https://github.com/ambient-intelligence-hq/egoproactive-verbalizer).