Staff ML Engineer: Agent Memory & Reasoning Systems

Socket.dev; United States; On-site; USD 180,000 - 240,000; Full time

Job summary

Wand is building a staff machine learning engineering team focused on agent memory, evolution, and reasoning. We are standing up a new University team where the work shifts away from training to deploying cloud-based, memory-driven agents.

For the next six months, there will be no model training; emphasis is on designing scalable memory systems, confidentiality, and measurable improvements. This is a senior, high-leverage role with broad impact across the org.

Qualifications

• Shipped production agents or agent-adjacent systems.
• Experience with memory, context engineering, or advanced reasoning without retraining.
• Applied, builder's mindset; deliverable in weeks, not semesters.
• Capable of owning ambiguous, senior-level problems.
• Strong software engineering fundamentals along with ML/agent experience.
• Familiar with vector databases and retrieval frameworks.
• Comfort with LLM provider APIs and embedding models.
• Experience with agent evaluation and benchmarking tools.
• Excellent written and verbal communication.

Responsibilities

• Build agent memory systems, generating and storing information.
• Design memory with confidentiality and scoping to prevent leaks.
• Develop systems to monitor agent behavior and extract best practices.
• Create reusable 'skills' for better reasoning and decision-making.
• Design tests and benchmarks to prove improvements.
• Help shape the technical roadmap for memory and reasoning.
• Tackle undefined problems and deliver shippable solutions.
• Document methodology for others to build on.

Skills

Agent memory

Context engineering

Memory systems design

Strong communicator

Education

MS or PhD

Tools

Pinecone

Weaviate

pgvector

LangChain

LlamaIndex

LangGraph

OpenAI API

Anthropic API

embedding models

Job description

Wand is building a staff machine learning engineering team focused on agent memory, evolution, and reasoning. We are standing up a new University team where the work shifts away from training to deploying cloud-based, memory-driven agents.

For the next six months, there will be no model training; emphasis is on designing scalable memory systems, confidentiality, and measurable improvements. This is a senior, high-leverage role with broad impact across the org.