Wand Synthesis AI Inc - Staff Machine Learning Engineer, Agent Memory & Reasoning (University)

United States, Hybrid3mo ago

In Office Staff NA Artificial Intelligence Higher Education Machine Learning Engineer Reporting pgvector Weaviate Pinecone Vector

Requirements

• You've shipped production agents or agent adjacent systems at a company, not just in a lab. • Experience with memory, context engineering, or techniques that make agents reason better without retraining them. • An applied, builder's mindset: rigorous thinking, shipped in days and weeks, not semesters. • Comfortable owning ambiguous, senior level problems on your own. • Strong software engineering fundamentals to go with your ML and agent experience. • Practical fluency with the modern agent tooling stack: vector databases (Pinecone, Weaviate, pgvector, or similar), retrieval frameworks (LangChain, LlamaIndex), and agent orchestration tools such as LangGraph. • Comfortable working directly with LLM provider APIs (OpenAI, Anthropic, or similar) and embedding models for retrieval and memory systems. • Experience with agent evaluation and benchmarking tooling (e.g. LangSmith, Ragas, TruLens, or a custom eval harness). • Strong communicator, written and verbal. • An advanced degree (MS or PhD), paired with real industry experience. • Experience testing and benchmarking agent behavior. • Experience building "skills" or reusable capabilities for AI agents. • Experience with agents that handle serious volumes of complex information (think a genuinely capable assistant, not a demo). • Time spent in a fast scaling product and engineering org.

Responsibilities

• Build agent memory systems: not just picking what goes into context, but the mechanisms that generate, curate, refine, and store that information in the first place. • Design memory with real constraints: confidentiality and scoping so agents never leak what they shouldn't. • Build systems that watch how agents behave across the org and turn that into shared best practices at scale. • Build reusable "skills" agents can call on: better reasoning, better financial decisions, better report writing. • Design and run tests and benchmarks that show whether these improvements actually work. • Help shape the technical roadmap for agent memory and reasoning as the team stands up. • Take an undefined problem and design a real, shippable solution for it. • Document your methodology clearly enough that others can build on it.