About Yodo Labs

Yodo Labs is an applied AI research and engineering company based in Tokyo and Osaka. We help enterprises take their most difficult AI projects from prototype to production. Our team includes PhDs and industry researchers working across the full AI stack.

About the Role

You believe that the hardest unsolved problems in AI aren't just about building bigger models — they're about making AI systems work reliably in the real world. You care about what happens after a demo works: Does the agent recover when a tool call fails? Does the retrieval pipeline hold up when the domain shifts? Can you trust the evaluation to catch regressions before users do?

At Yodo Labs, our research is driven by questions that surface from production. We work at the boundary between what frontier models can do in theory and what they actually do when deployed at scale for enterprise clients. As a Research Scientist, you will own the cycle from identifying these gaps, to designing experiments, to publishing results that feed back into our systems.

This is not a pure-paper role. You will prototype alongside engineers, ship alongside product, and write alongside the research community. We view research and engineering as two sides of the same coin — every researcher on our team writes production code, and every engineer reads papers.

What You'll Do

Define and own research questions that emerge from real production challenges — not assigned top-down, but identified from what's breaking or underperforming
Design rigorous experiments, run them quickly, and iterate — we prioritize learning speed over publication polish
Collaborate with engineers to move research findings into production systems, often within weeks
Publish results externally where they advance the field, and internally where they inform decisions
Contribute to our evaluation infrastructure — we believe measurement is research, not overhead

Representative projects

Designing a benchmark to measure multi-step agent reliability across different failure modes (tool errors, context loss, hallucinated actions) and using the results to improve our agent framework
Investigating why a RAG pipeline performs well on English benchmarks but degrades on Japanese enterprise documents, and proposing architectural changes
Prototyping a lightweight evaluation harness that lets clients test LLM quality on their own domain-specific tasks without ML expertise
Running controlled experiments comparing fine-tuning vs. in-context learning for structured data extraction, and writing up findings for publication

You may be a good fit if you have

A research background (PhD, Master's, or equivalent experience through significant projects and publications)
Strong implementation skills — you can go from hypothesis to working experiment in a day, not a month
Experience with LLMs, retrieval-augmented generation, agents, or evaluation methodology
A preference for applied, messy problems over clean theoretical ones — production AI is full of surprises
The instinct to ask "does this actually work?" rather than "does this look good on a benchmark?"
Comfort working in a small team where you'll pair with engineers, talk to clients, and context-switch between research and shipping
Business-level English; conversational Japanese is helpful but not required

Candidates need not have

A PhD — strong research output matters more than credentials
Experience in a specific subfield — we care about research taste and engineering ability, not topic match
Prior industry experience — some of our best work comes from people straight out of academia who are hungry to see their ideas deployed

How to Apply

Send an email to team@yodolabs.jp with the subject line "Research Scientist (Full-time) — [Your Name]". Please include:

Your CV or resume
A brief note on one research problem you'd want to explore at Yodo Labs — we're more interested in how you think about problems than which problem you pick
Links to relevant publications, preprints, or project repositories

Logistics

Location: Tokyo / Osaka, Japan — remote-friendly Work style: Flexible hours; we value output over presence Visa sponsorship: We sponsor work visas for Japan where possible

We encourage you to apply even if you don't meet every qualification listed.