Memory Systems AI Research Engineer

Mem0; San Francisco (CA); On-site; USD 175,000 - 250,000; Full time

Benefits offered by this job

Health coverage

Dental coverage

Vision coverage

In-office meals (Lunch/Dinner)

Flexible PTO

Equity

Team events

Quality equipment

Job summary

Mem0, based in San Francisco, is seeking a research-engineering leader to own memory features end‑to‑end. You’ll fine‑tune models for extraction, updates, consolidation/forgetting, and dreaming; turn customer pain points into research hypotheses; implement and benchmark ideas; and ship them with high‑quality latency and cost in collaboration with Engineering.

You’ll build evaluation at scale, create offline metrics and online A/B tests, and close the loop with production feedback to keep pushing

Qualifications

• Experience in RAG or information retrieval for real products.
• Model training/fine-tuning with LLMs and encoders.
• Strong Python; PyTorch; familiarity with vLLM and serving frameworks.
• Built evaluation for language retrieval and generation (offline/online).
• Orchestrate data pipelines for production with low-latency SLAs.

Responsibilities

• Train models for memory extraction, updates, and forgetting.
• Prototype ideas from papers and productionize winning approaches.
• Build evaluation layer with metrics, gold sets, and dashboards.
• Identify customer pain points and validate through field trials.
• Design APIs and data contracts; plan safe rollouts with engineering.

Skills

RAG/IR experience

Python

PyTorch

vLLM

LLM serving frameworks

Tools

Vector databases

CI/CD tooling

Job description

Mem0, based in San Francisco, is seeking a research-engineering leader to own memory features end‑to‑end. You’ll fine‑tune models for extraction, updates, consolidation/forgetting, and dreaming; turn customer pain points into research hypotheses; implement and benchmark ideas; and ship them with high‑quality latency and cost in collaboration with Engineering.

You’ll build evaluation at scale, create offline metrics and online A/B tests, and close the loop with production feedback to keep pushing