AI Research Engineer – Generative AI

We’re seeking an AI Research Engineer to develop multimodal LLMs, reasoning agents, and large‑scale RAG systems. In this role, your work will directly help people living with neurological disease; you’ll be part of a fast-moving group dedicated to using AI tools to accelerate impact of an FDA-approved implantable device. You’ll own model training pipelines, high‑throughput data and indexing systems, and retrieval‑augmented generation powering production applications. Ideal candidates have experience training large multimodal models, building scalable data/feature stores, and shipping agentic systems with strong observability and safety. The position must be based in the Bay Area and with the ability to commute to our NeuroPace Mountain View office three days a week.

Key Responsibilities

Model Training & Inference 

• Train/fine‑tune LLMs and multimodal models (text/vision/audio; bonus: neural/ECoG). 
• Use SFT, DPO/ORPO, RLHF/RLAIF, and parameter‑efficient tuning (LoRA/QLoRA). 
• Build agentic systems (tool use, planning, memory, reasoning) with attention to latency, safety, and cost. 
• Optimize inference: quantization, graph/tensor optimizations, speculative decoding, distillation, caching. 

Data & Indexing 

• Build fast pipelines for ingest, feature extraction, multimodal alignment, and dataset curation. 
• Design billion‑scale indexing (vector, hybrid sparse+dense, graph‑based). 
• Implement observability: lineage, drift, coverage, bias, differential performance, cost. 

RAG & Knowledge Systems 

• Build production RAG (chunking, hybrid search, reranking, context optimization). 
• Maintain continuous refresh pipelines (incremental indexing, TTL, schema evolution). 
• Integrate ontologies/knowledge graphs for grounding and attribution. 

Evaluation, Safety & Reliability 

• Develop structured evaluations for faithfulness, factuality, consistency, and multimodal alignment. 
• Implement guardrails: prompt hardening, policy filters, PII/PHI handling, output verification. 
• Run A/B tests; define metrics for quality, latency, and cost. 

Platform & Collaboration 

• Partner with infra for distributed training (FSDP/ZeRO, multi‑GPU/TPU), registries, and ML CI/CD. 
• Mentor engineers and contribute to long‑term technical strategy. 

Requirements

Minimum Qualifications 

• 2+ years in ML/AI (or equivalent impact). 
• Master’s in CS or related field, or equivalent experience. 
• Hands‑on LLM/multimodal training (PyTorch/TensorFlow/HuggingFace/SageMaker). 
• Strong engineering skills: Python, TypeScript, APIs, testing, profiling, and observability. 

Preferred Qualifications 

• 4+ years in ML/AI and production ML system experience. 
• Experience shipping agentic systems with planning, tool use, schemas, memory, and guardrails. 
• Multimodal expertise: CLIP‑style contrastive models, BLIP/LLaVA/Flamingo, Whisper/NeMo, video, neural signals. 
• Production RAG (chunking, grounding, hallucination mitigation, cost/latency tuning). 
• Distributed training (FSDP/DeepSpeed/ZeRO), mixed precision, checkpointing. 
• Information retrieval: query understanding, retrieval‑aware generation, knowledge graphs. 
• Causal inference or counterfactual evaluation for scientific/biomedical ML. 

$135K-$175K Base - Compensation will be determined based on several factors, including but not limited to skill set, years of experience, and geographic location.

NeuroPace is proud to be an equal opportunity employer and values the contributions of our culturally diverse workforce.San Francisco and Los Angeles applicants: The Company will consider for employment qualified applicants with Criminal Histories in a manner consistent with the requirements of the Los Angeles Fair Chance in Hiring Ordinance or the San Francisco Fair Chance Ordinance (as applicable)