AI Agent Engineer @ Roe (YC F24) | San Francisco Bay Area

I develop and deploy production AI agents operating in enterprise environments.

Separately, I explore how learning systems can be trained more like humans - through feedback, revision, and structured reasoning - grounded in quantitative/abstract methods.

• Multi-step, Temporal-orchestrated agent systems
• Compliance, AML, and risk automation
• Audit-first, artifact-driven architecture
• First-principles approach to learning, reasoning, and system design

At Work

• Agent orchestration & long-running workflows
• Structured extraction & schema-constrained generation
• Risk intelligence & adverse media detection
• Event-driven automation

Independent Research

• Credit assignment for multi-step reasoning trajectories
• Learning via critique, revision, and feedback loops (human-like learning)
• Dataset design for reasoning (trajectory, preference, and process data)
• Evaluation of reasoning quality, not just outcomes

LLMs: Tool-using agents with context management, evaluations on large and small models

Infra: AWS, EKS, Vercel, Terraform, Kubernetes

Data: ClickHouse, AWS stack

Research: Python, JAX, PyTorch, statistical modeling, deep quantitative methods

I study how to formalize learning processes observed in humans:

• Iterative refinement (attempt → critique → revision)
• Credit assignment across sequences, not just final outcomes
• Balancing exploration vs. correctness under uncertainty
• Structuring training signals using statistical and econometric principles

I approach this from a quantitative POV - thinking in terms of signal, noise, bias, and identifiability—rather than heuristics - if there is ever a way I can borrow from another field such as physics, economics, neuroscience, or mathematics I take that opportunity.

LinkedIn: https://linkedin.com/in/jadenfix Location: San Francisco Bay Area

Build agents that learn the way people do—through iteration, feedback, and structure.