ML Engineer: Memory Systems (In-Person, SF, 6 mo)

Engramme; San Francisco (CA); On-site; USD 40,000 - 60,000; Part time

Job summary

Engramme, a Frontier AI lab spun out of Harvard, seeks a 6-month in-person engineer to build automated eval pipelines for memory retrieval and reconstruction in San Francisco. You will design benchmarks, analyze model behavior, and contribute to production-grade evaluation systems alongside backend and research teams.

The role emphasizes strong Python and ML framework experience, plus background in eval pipelines and benchmarking. Startup experience and industry tooling familiarity are a plus.

Qualifications

• Strong Python and ML framework experience.
• Background in eval pipelines, model testing, or benchmarking.
• Understanding of LLMs, embeddings, or retrieval systems.
• Ability to analyze failures, define metrics, and reason from first principles.
• Comfortable moving fast and owning core architecture.

Responsibilities

• Build automated eval pipelines for memory retrieval, compression, and reconstruction.
• Design benchmarks for context retention, long-term consistency, and memory safety.
• Analyze model behavior, surface failure modes, and define quality metrics.
• Run large-scale experiments and maintain continuous evaluation systems.
• Develop internal tools, dashboards, and workflows for fast debugging.
• Integrate evals directly into production with backend + research teams.

Skills

Python

ML frameworks

eval pipelines

LLMs/embeddings

fault analysis & metrics

Job description

Engramme, a Frontier AI lab spun out of Harvard, seeks a 6-month in-person engineer to build automated eval pipelines for memory retrieval and reconstruction in San Francisco. You will design benchmarks, analyze model behavior, and contribute to production-grade evaluation systems alongside backend and research teams.

The role emphasizes strong Python and ML framework experience, plus background in eval pipelines and benchmarking. Startup experience and industry tooling familiarity are a plus.