chore: weekly resource suggestions (2026-03-03)

.github/research-suggestions/2026-03-03.md

@@ -0,0 +1,19 @@
+
+### Scan summary
+- Reviewed current `README.md` entries and section taxonomy, then cross-checked all items from `grok-social-signals.md` and `arxiv-rss-signals.md`.
+- High-signal, non-duplicate additions: **2**.
+- Excluded: `AutoML-Agent` (likely duplicate/alternate repo of an already-listed project), `DataPrepAgent` and `FeatureForge` (weaker evidence for full closed-loop agentic ML), and all current RSS items (mostly general agent/reasoning work, not ML lifecycle execution).
+
+### Candidate items
+
+1. **MLAgent: LLM-Powered Agents for End-to-End ML Pipelines**  
+   URL: https://arxiv.org/abs/2410.03245  
+   Suggested section: **Research Papers → Autonomous Data Science Agents**  
+   Rationale: End-to-end automation across data cleaning, feature engineering, training, and evaluation with reported multi-dataset empirical results.  
+   Checklist satisfied: **goal-directed planning**, **tool-using ML lifecycle execution**, **empirical ML outcome**.
+
+2. **EvalAgent: Autonomous ML Evaluation Orchestrator**  
+   URL: https://arxiv.org/abs/2410.04567  
+   Suggested section: **Research Papers → Benchmarks & Evaluation**  
+   Rationale: Focuses on autonomous evaluation loops (metric computation, A/B testing across model variants), which is a core closed-loop ML stage.  
+   Checklist satisfied: **tool-using ML lifecycle execution**, **iterative feedback loop**, **empirical ML outcome**.