Who you'll join

A pre-seed AI startup building a taste layer for AI models. The founders are teaching AI to understand subjective domains like design, writing, personality, and visual style. They closed a $2M angel round and are moving fast toward product-market fit with pilots already running at major AI labs.

What you'll do

1. Train reward models and classifiers that evaluate subjective qualities in text, images, and design.
2. Build evaluation frameworks and benchmarks for domains where objective metrics fail.
3. Run post-training experiments on open-source models to test new data formats and techniques.
4. Work directly with AI lab researchers and creative domain experts to design and execute pilots.
5. Own end-to-end research pipelines from data preparation through results publication.
6. Publish blogs and whitepapers documenting findings and methods.
7. Move between one-week iteration cycles and multi-month research projects.

Who you are

1. You have hands-on experience training and fine-tuning large language or diffusion models through post-training techniques like SFT, RLHF, or DPO.
2. You think like a researcher but move like an engineer. You iterate fast, test ideas quickly, and ship results.
3. You've worked at a data company, with data vendors, or on post-training infrastructure. You understand data pipelines and quality deeply.
4. You care intellectually about the problem space. A personal interest in design, writing, art, or creative fields is a strong signal.
5. You're comfortable with ambiguity and early-stage constraints. You've either worked in startups or understand what it takes to build one.
6. You have low ego and collaborate well. You're excited to work alongside a small, tight team.
7. Bonus: Experience with multimodal models or background from companies like Pika, Luma, Midjourney, or Black Forest Labs.

Tech stack

Python, PyTorch, LLMs, diffusion models, post-training frameworks, data pipelines.

Why you'll thrive

1. The problem is genuinely hard and intellectually stimulating. You're teaching AI what subjective quality means across domains nobody else is tackling at this scale.
2. Compensation is $200K–$350K with 0.25–1.5% equity depending on experience. The founders flex on both for the right senior candidate.
3. You ship fast and see impact immediately. Research cycles run one week to one quarter, not months in a vacuum.
4. The team actively publishes findings and talks publicly about the work. Your research gets visibility and credit.
5. You join at the exact right moment. The startup has resources to grow aggressively, pilots running with major labs, and clear revenue visibility within months.