Member of Technical Staff (Machine Learning Research Engineer)
Apply at source. Perplexity handles the application directly; Houtini doesn't take a fee from candidates or companies. We curate which companies appear; the listings come from yubhub.
What the team is looking for.
We are seeking an experienced Machine Learning Research Engineer to help build the next generation of advanced search technologies, with a focus on retrieval and ranking.
As a key member of our Search team, you will be responsible for pushing search quality forward through models, data, tools, or any other leverage available. This includes architecting and building core components of the search platform and model stack, designing, training, and optimizing large-scale deep learning models using frameworks like PyTorch, and deploying models in a scalable and performant way.
Responsibilities:
- Relentlessly push search quality forward , through models, data, tools, or any other leverage available
- Architect and build core components of the search platform and model stack
- Design, train, and optimize large-scale deep learning models using frameworks like PyTorch, leveraging distributed training (e.g., PyTorch Distributed, DeepSpeed, FSDP) and hardware acceleration, with a focus on retrieval and ranking models
- Conduct advanced research in representation learning, including contrastive learning, multilingual, and multimodal modeling for search and retrieval
- Deploy models , from boosting algorithms to LLMs , in a scalable and performant way
- Build and optimize RAG pipelines for grounding and answer generation
- Collaborate with Data, AI, Infrastructure, and Product teams to ensure fast and high-quality delivery
Qualifications:
- Deep understanding of search and retrieval systems, including quality evaluation principles and metrics
- Proven track record with large-scale search or recommender systems
- Strong proficiency with PyTorch, including experience in distributed training techniques and performance optimization for large models
- Expertise in representation learning, including contrastive learning and embedding space alignment for multilingual and multimodal applications
- Strong publication record in AI/ML conferences or workshops (e.g., NeurIPS, ICML, ICLR, ACL, CVPR, SIGIR)
- Self-driven, with a strong sense of ownership and execution
- Minimum of 3 years (preferably 5+) working on search, recommender systems, or closely related research areas
- PyTorch
- Deep learning
- Representation learning
- Contrastive learning
- Multilingual modeling
- Multimodal modeling
- Search and retrieval systems
- Quality evaluation principles and metrics
- Distributed training techniques
- Performance optimization for large models
Other roles you might consider.
Filtered through the same AI-companies allowlist.
Technical Program Manager (TPM), Infrastructure
Cursor
Production Manager
ElevenLabs
Production Manager
ElevenLabs
Data Center Energy Lead, Australia
Anthropic
Software Engineer, Ads Product
xAI
Lead, Operations & Maintenance (O&M)
xAI
New to AI work? Start with these.
Six pieces of orientation. Most AI-company job specs assume you've done this kind of hands-on work already. If you haven't, an afternoon with one of these is the cheapest way to close the gap.
Claude Desktop, from zero.
The agentic-AI assistant most of the people you'd be working alongside use every day. Install, configure, first useful prompts.
What MCPs areThe best MCPs for Claude Desktop.
MCP servers extend an AI assistant with tools and data. The catalogue most teams use. Useful technical context for any AI-engineering role.
Code with AIClaude Code, the complete beginners' guide.
The CLI for AI-paired development. Required reading if you're applying for any engineering role that mentions agents, or any role full stop.
Run a local modelHow to set up LM Studio.
Running a model on your own machine teaches you more about how AI products work in three hours than a year of using ChatGPT will.
The hardware realityBeginner's guide to AI hardware.
What the infrastructure under the model actually looks like. Useful context for infrastructure, applied-AI and hardware roles.
Browse the stackMCP catalogue.
Eleven MCP servers Houtini maintains or recommends. Each detail page describes a real piece of working AI infrastructure.