I research how AI agents should handle human context and memory. I build AI systems, from zero-to-one prototypes to globally scaled products, across applied ML, evaluation, and data infrastructure.

Right now I'm at Recurse Center, exploring agentic systems, RLVR, and leading a DDIA study group.

• Memory architectures for agents: How should AI agents handle personal context? I'm exploring memory systems grounded in cognitive psychology. Paper under review
• Political polarization through language: Measuring political polarization through semantic divergence across five Spanish parliamentary legislatures

• Adaptive Synth Data: Self-improving data generation system using bandit-based strategy selection
• RAG Search Engine: Vector search engine with ANN indexing built from first principles
• Fact-Checking NLP: Semantic similarity pipeline for automated claim verification in Spanish political discourse

• Proof of Capture: Cryptographic media provenance for photos. WIP
• iClaude: Autonomous agent that controls iOS devices through screen observation and accessibility metadata
• Find My Cenote: Open dataset mapping cenotes in the Yucatán, enriched with remote sensing and LiDAR features

• Kinship: Founding AI Research Engineer. Architected a knowledge graph synthesizing multimodal context into a personal memory interface
• Apple: ML Engineer on the Siri team. Led the international data strategy for Siri. Built LLM-based evaluation systems. Part of the core Siri expansion for Vision Pro
• Newtral: Built the core ML infrastructure for automated fact-checking from scratch
• Financial Times: FT Labs intern, web components and accessibility tooling

hi@merybenavente.me · merybenavente.me · projects · LinkedIn · X