AI Safety Researcher · Systems Builder · AAAI 2026 First Author
I build high-leverage systems across domains. Started in biotech, moved through industrial engineering and entrepreneurship, ended up in AI research — not by chasing titles, but by following meaningful problems. Based in Trinidad and Tobago.
View Work ↓AAAI 2026 workshop paper. Can you safely use a powerful but untrusted AI to plan tasks, while smaller trusted AIs execute the work? Yes — but only if you monitor the execution, not just the plan. Implementation-level monitoring achieves AUROC 0.96 vs. 0.52 for plan-only monitoring.
When an AI is about to deceive you, does it show physical signs — like a person who pauses before lying? DeceptionScoop tests this: measuring GPU power spikes, response latency, and token entropy shifts during LLM inference to detect deceptive intent before it acts.
Public substrate for foundational ontological primitives. Four-layer architecture: base layer (provenance, identity, assertion, signature), structural layer (relation, taxonomy, context, delegation), real-world benchmarks (66% token efficiency, 94-97% cache hit rates), and patterns for safe autonomous systems. Enables agents and humans to name things durably and reason together with semantics that outlast implementations.
A personal AI operating system. Persistent memory across sessions, multi-channel presence (Discord, Telegram, Signal), ontology-driven context engine, and autonomous background operations. Levi is the user-facing interface layer of a broader intelligent systems stack — purpose-built to know who you are, what you're working on, and what needs to happen next.
Enterprise intelligence platform for capital-intensive industries. Ontology-driven digital twin engine for proposal automation, asset tracking, and supply chain forecasting. Talos connects instrumentation data with structured AI reasoning — designed for industries where decisions are expensive and information is fragmented across systems, teams, and time.
Financial intelligence OS. Tracks capital flow, smart money movement, and market structure across DeFi and traditional markets. Flux is built to understand where money actually moves before narratives form — mapping liquidity patterns, on-chain flows, and structural positioning. Not a trading bot. A financial lens for people who think in systems.
Foundational operational intelligence layer — ontology-driven infrastructure for context management, agent orchestration, and decision support across systems. Node Zero is the substrate beneath Talos, Levi, and Flux: the layer that holds structure, propagates context, and coordinates autonomous components across the full stack.
POP2 is a human enzyme involved in RNA processing. We used AI throughout the workflow to help bridge technical gaps: supporting literature review, lab calculations, protocol refinement, parameter selection, experimental design, and troubleshooting. The result was 735 μg/mL expression yield in yeast and E. coli, with a 10-person undergraduate team operating at graduate-level throughput. CURC Highest Honors.
Computational R&D pipeline for cryoprotectant discovery. Curated molecular database of 9,400+ compounds with RDKit-powered similarity search, cheminformatics scoring, and property filtering. Built for a Colorado biotech lab. Archived R&D with future potential — core dataset and codebase remain private.
Service engineering internship in Trinidad. Designed P&ID diagrams for automated brewery control in AutoCAD, implemented CCTV-PLC monitoring for regulatory compliance, and supported industrial site work tied to process and infrastructure operations.
Founded the first specialty retail smoke shop in Trinidad by studying U.S. 2018 Farm Bill regulations and mapping them onto local legal gray zones. Ran operations solo: supplier sourcing, compliance, inventory, customer acquisition. Achieved traction, executed a clean exit, and relocated to pursue research.
Building Levi and Talos. Working on systems for memory, context, operational intelligence, and decision support. Ongoing research collaboration targeting publication in Q3 2026.
I started in fermentation science at Colorado State University, then moved wherever the most meaningful problems seemed to be. That path ran through biotech, industrial automation, entrepreneurship, and AI research. I'm the first author of Factor(UT), accepted at AAAI 2026. I've built cryoprotectant discovery pipelines, designed P&ID systems for industrial control, and built systems for memory, context, and agent orchestration. The throughline isn't a title — it's a bias toward high-impact work and a way of thinking across domains.
Single-domain expertise is valuable, but increasingly compressible. What's harder to replace is someone who can move between biotech, financial systems, industrial automation, and AI — and actually connect them. Problems look different from each vantage point. The cross-domain connections are often where the leverage is.
I build systems that reflect that style of thinking: modular, composable, grounded in memory and context. Not just tools for a task — infrastructure for working on problems that don't fit neatly into one box.