Juan Miguel Navarro Carranza

Software Engineer

Google

2026-10 -- Present

Software engineering at Google.

AI Research Engineer

Daice Labs · Boston, U.S.

2026-03 -- 2026-09

Designing hybrid AI systems that integrate foundation models with neurosymbolic reasoning and biologically inspired computational principles. Developing robustness evaluation and governance tooling to audit system behavior under contextual variation, and collaborating across engineering and product teams to translate research prototypes into scalable, production-ready AI solutions.

• Prototyping hybrid composite AI architectures that combine foundation models with neurosymbolic components and bio-inspired system design for adaptive, auditable workflows.
• Developing evaluation and governance-minded tooling to stress-test hybrid systems under changing context and to support traceable, explainable decision-making.
• Collaborating across engineering and product teams to translate research findings into production-ready AI solutions.

Graduate Researcher

Stanford University · Stanford, U.S.

2019 -- 2025

• Developed and published a reproducible evaluation protocol (NeurIPS 2025 Workshop) measuring surface-form robustness of LLMs via paraphrase stress tests, using Mistral-7B and Qwen2.5-7B with deterministic 4-bit inference on NVIDIA A100.
• Designed and implemented large-scale HPC pipelines for nonlinear dynamic simulations of reinforced concrete buildings, scaling to hundreds of thousands of earthquake simulations on Stanford’s Sherlock cluster.
• Built a modular Python data ecosystem integrating DuckDB (relational metadata) and HDF5 (time-series arrays), enabling fast querying and analysis of multi-terabyte structural simulation datasets.
• Parametric calibration of the Reinforcement Ductile Fracture Model (RDFM), integrating mechanics and ML to predict reinforcement fracture under low-cycle fatigue of reinforced concrete shear walls; validated with 23 laboratory experiments.
• Applied machine learning methods (Bayesian inference, probabilistic risk modeling) to quantify collapse fragility, resilience metrics, and post-earthquake degradation of building systems.
• Produced visualization and analytics tools (Matplotlib, Jupyter, HPC logging) for real-time monitoring and reproducibility of simulations across thousands of jobs.