Teaching Agents to Search with NVIDIA Data Designer — Dhruv Nathawani, NVIDIA
Dhruv Nathawani uses that funnel to explain how NVIDIA builds synthetic data that teaches a model to search, rather than answer from memory.
Proof1 independent outlet
Key points
- 50,000 seed examples become 7,000 training records after generation and filtering.
- Data Designer turns those seeds into questions, tool calls and responses, with Tavily providing search through MCP.
- Questions distinguish seed generation from agent execution, explore reinforcement learning alternatives and discuss other training tasks.
- 0:00 - Turn an agent behavior into a data pipeline
Sources (1)
- [1]Teaching Agents to Search with NVIDIA Data Designer — Dhruv Nathawani, NVIDIAAI Engineer (YouTube) · Oct 11, 04:30 PM
Dhruv Nathawani uses that funnel to explain how NVIDIA builds synthetic data that teaches a model to search, rather than answer from memory.
50,000 seed examples become 7,000 training records after generation and filtering.
Extractive summary: sentences quoted from the sources.
Before this
- Oct 11, 2026Stereo-Planner: Learning Stereo-to-Trajectory Planning for Zero-Shot Sim-to-Real Quadrotor Navigation
- Oct 10, 2026TIDE: Trajectory-Informed Skill Distillation for Efficient Text-to-Video Prompt Optimization
- Oct 10, 2026RES-MR: Risk-Aware Reasoning for Explainable and Safe Medication Recommendation
- Oct 9, 2026Minions: A Dynamic and Distributed Virtual GPU Cloud Built on Idle Apple Silicon Devices
- Oct 9, 2026Continually Post-training Small Models via Synthetic Self-distillation
- Oct 8, 2026StereoEngine: Scaling Stereo Bimanual Robot Data without Stereo Cameras