Elliot Slusky

AI / ML Researcher

Miami, FL elliot@slusky.com LinkedIn

I am an AI and machine learning researcher focused on large language model efficiency, model routing, agentic systems, and latent reasoning. At Hazy Research at Stanford University, I study the energy and compute trade-offs of frontier model inference and develop routing architectures that optimize intelligence per joule and per dollar.

Alongside academic research, I have built evaluation harnesses and fine-tuning pipelines for production agents at an a16z-backed startup, conducted security and intelligence analysis at the Human Rights Foundation, and previously led a software engineering consultancy that generated over $10M in client revenue.

Selected Research

Submitted to ICLR

The Intelligence Control Plane: Optimizing Intelligence per Joule and per Dollar for Agentic Tasks

Hazy Research · Stanford University — Investigated AI model energy consumption and inference efficiency across agentic workloads, developing query-routing strategies that reduce energy usage by up to 80% while preserving task performance.

Experience

Research Assistant · Hazy Research · Stanford University

  • Researched AI model energy and efficiency; developed strategies to save 80% of energy through routing queries to the most appropriate model.
  • Submitted to ICLR a paper called "The Intelligence Control Plane: Optimizing Intelligence per Joule and per Dollar for Agentic Tasks".
LLM Inference Energy Efficiency Model Routing ICLR

Part-Time AI Developer · Meridian.AI (Stealth Startup, backed by a16z)

  • Built evaluation suites to benchmark and systematically improve agent performance.
  • Developed a data ingestion platform for fine-tuning foundation models.
  • Instrumented client-side observability to monitor agent behavior in production.
Agent Evals Fine-Tuning Observability

Independent Researcher · Self-Directed

  • Researched latent reasoning mechanisms in language models and implemented a recurrent looped Transformer architecture inspired by recent work on adaptive computation.
  • Experimented with Mixture-of-Experts (MoE) models to detect and prune underutilized experts, increasing inference efficiency and reducing memory footprint.
Latent Reasoning Looped Transformers Mixture-of-Experts

AI Research Intern · Human Rights Foundation

  • Analyzed data on the Venezuelan military to identify key structural, strategic, and personnel insights.
  • Built AI-powered vetting tools to support security and due-diligence workflows.
  • Contributed to systems for rapid analysis of authoritarian regimes.
Intelligence Analysis Security Vetting Applied AI

Founder & Lead Developer · Web3 Development Consulting · San Francisco

  • Managed a team of three developers building and shipping production software for clients.
  • Generated over $10 million in client revenue through applications deployed on Solana.
  • Engineered systems to interact with and trade decentralized financial markets across Ethereum and Solana.
Solana Ethereum Distributed Systems

Software Engineering Intern · Cookbook.dev · San Francisco

  • Built minimum viable products (MVPs) and integrated them into production.
  • Contributed to Cookbook's smart contract library.
  • Created ChefGPT, one of Cookbook's most widely adopted features with over 5,000 monthly active users.
LLM Applications Smart Contracts Product Engineering

Hackathons & Awards

1st Place

AGI Fight Club · NFX

Winner of a four-round timed engineering competition building applied AI systems.

3rd Place (of 50 Teams)

Cognition / OpenAI Hackathon

Built Voice Of The People, a real-time system for analyzing political sentiment at scale.

Technical Areas

Machine Learning & Research

Large Language Models Inference Efficiency Model Routing Mixture-of-Experts Latent Reasoning Fine-Tuning & Evals

Languages & Frameworks

Python Rust JavaScript / TypeScript PyTorch Agentic Frameworks

Data & Systems

Vector Databases Data Analysis Multi-GPU Pipelines Production Observability