my work

My work spans interpretability and controllability, generative world modeling, and visual analytics / HCI, all grounded in a broader goal: building AI systems that are emergent, intuitive to use, and robust in real-world environments. I focus on improving multimodal representations and reasoning (text, 2D, 3D, video, action), and on designing optimization strategies that guide model behavior in predictable, controllable ways.

Recently, I’ve focused on improving LLM reasoning, alignment, and tool use—leading the design of post-training and agentic AI pipelines (including RLVR-based methods) that support long-horizon, multi-step capabilities. My work combines research with product impact: systems I build ship in production and support real users and teams.

At Apple, I’ve shipped research-driven systems that power real product experiences:

• Interpretability & Controllability — developed data-centric observability and experiment tracking tools now used across Siri and Vision Pro ML algorithms
• 3D/4D Generation — prototyped and scaled on-device text-to-image-to-3D generation models, supporting content-generation workflows for Apple Vision Pro.
• Agentic AI — drove post-training, RL, and RLVR initiatives for tool use, alignment, and long-horizon reasoning, informing the design of next-generation agentic Siri systems.

more background (toggle) 👈

publications

• Miao, K., Agrawal, H., Zhang, Q., Semeraro, F., Cavallo, M., Gu, J., and Toshev, A. (2024). DSplats: 3D Generation by Denoising Splats-Based Multiview Diffusion Models.
• Zhang, Q., Zhai, S., Bautista, M. A., Miao, K., Toshev, A., Susskind, J., and Gu, J. (2024). World-consistent Video Diffusion with Explicit 3D Modeling.
• Yuan, J., Miao, K., Walker, I., Oh, H., Katolikyan, T., Xue, Z., ... and Cavallo, M. (2024). VIBE: A Visual Analytics Workflow for Subgroup-based Semantic Error Analysis of CVML Models.
• Lin, T., Yuan, J., Miao, K., Katolikyan, T., Walker, I., and Cavallo, M. (2024). XR VIS: Designing Visual Analytics for 3D CVML Model Debugging Across XR Spectrum.
• Miao, K., Gokul, A., Singh, R., Petryk, S., Gonzalez, J., Keutzer, K., Darrell, T., and Reed, C. (2022). Knowledge-Guided Self-Supervised Vision Transformers for Medical Imaging.
• Miao, K., Gokul, A., Singh, R., Petryk, S., Gonzalez, J., Keutzer, K., Darrell, T., and Reed, C. (2022). Prior Knowledge-Guided Attention in Self-Supervised Vision Transformers.
• Miao, K., Friesner, I., Dahle, J., Yousefi, S., Buchake, B., Kaur, P., Odisho, A. Y., Cinar, P., and Hong, J. C. (2021). Machine learning-based approach to assessing risk of outpatient cancer treatment-related emergency care.
• Friesner, I. D., Miao, K., Dahle, J., Zack, T., Feng, J., Yousefi, S., Buchake, B., Kaur, P., Cinar, P., Kidder, W. A., Odisho, A. Y., and Hong, J. C. (2023). Prospective validation of machine learning-based approaches to predict potentially preventable emergency visits and hospitalizations.
• Matsunaga, T., Reisenman, C. E., Goldman-Huertas, B., Brand, P., Miao, K., Suzuki, H. C., Verster, K. I., Ramírez, S. R., and Whiteman, N. K. (2019). Evolution of olfactory receptors tuned to mustard oils in herbivorous Drosophilidae. Molecular Biology and Evolution.