Research Scientist

techire ai; San Francisco (CA); On-site; USD 250,000 - 400,000; Full time

Job summary

techire ai in San Francisco is building frontier reasoning models for scientific discovery, seeking researchers who own complex reasoning systems from data to production. You will design training pipelines, develop reasoning datasets, and evaluate multi-step hypotheses.

Ideal candidates have hands-on experience with LLM post-training, reinforcement learning for reasoning, and evaluation or alignment, and prefer multimodal or vision-language contexts.

Qualifications

• Ideal for researchers who have built reasoning systems themselves.
• Strong experience in LLM post-training, RL for reasoning, reasoning datasets, evaluation or alignment is priority.
• Experience with multimodal or vision-language models (VLMs) is attractive.

Responsibilities

• Developing reasoning models and post-training pipelines for frontier language and multimodal models.
• Building evaluation frameworks and benchmarks for complex, multi-step reasoning.
• Improving model performance through reinforcement learning, alignment and scalable experimentation.
• Exploring reasoning across multimodal inputs, where language and vision intersect to solve scientific problems.
• Working closely with a small research team to rapidly test and deploy new ideas.

Skills

Reasoning systems

Post-training

Reinforcement learning

Evaluation benchmarks

Multimodal models

Tools

Benchmark frameworks

Training pipelines

Job description

Interested in building reasoning systems that push AI beyond prediction?

You'll join a well-funded, stealth AI start-up building frontier reasoning models for scientific discovery. The team is tackling problems where models need to generate, evaluate and refine hypotheses across complex, multi-step scientific workflows.

It's an opportunity to work on genuinely novel AI for Science research, combining frontier reasoning, post-training and reinforcement learning with a proprietary dataset that creates a significant competitive advantage.

You'll work across the full research lifecycle, from designing training pipelines and reasoning datasets through to evaluation, experimentation and production systems. The emphasis is on building models that reason, not just models that respond.

Your work will include:

• Developing reasoning models and post-training pipelines for frontier language and multimodal models.
• Building evaluation frameworks and benchmarks for complex, multi-step reasoning.
• Improving model performance through reinforcement learning, alignment and scalable experimentation.
• Exploring reasoning across multimodal inputs, where language and vision intersect to solve scientific problems.
• Working closely with a small research team to rapidly test and deploy new ideas.

This role is ideal for researchers who have built reasoning systems themselves. Strong experience in LLM post-training, RL for reasoning, reasoning datasets, evaluation or alignment is the priority.

Experience with multimodal or vision-language models (VLMs) would also be attractive, especially where you've worked on reasoning, although exceptional researchers from pure language model backgrounds are equally encouraged to apply.

You could be an exceptional PhD graduate, Research Engineer, Applied Researcher or experienced Research Scientist. What matters most is hands-on ownership of reasoning systems rather than simply working alongside them.

$250K - $400K Base (D.O.E.) + equity + very lucrative bonus/success structure

If you're looking for an opportunity to work on frontier reasoning research in a small, ambitious team where your work directly shapes both the technology and the direction of the company, we'd love to hear from you.