Product Engineer

HorizonAI Talent; San Francisco (CA); On-site; USD 150,000 - 300,000; Full time

Job summary

HorizonAI Talent in Palo Alto seeks a Product Engineer (AI) to own end-to-end features for a seed-stage LegalTech AI product. You’ll design the agentic pipeline, build the backend, and craft the React interface the customer touches.

You’ll own evals and observability, ensuring AI outputs are reliable and regressions are caught before customers see them. Expect hands-on work with n8n, OpenAI agent builder, Claude SDK, and strong Python with PyTorch/TensorFlow/HuggingFace.

Qualifications

• 2+ years in engineering with AI/ML, LLMs and evaluations.
• Agentic workflows built and evaluated in production, across backend and frontend.
• Hands-on with agent frameworks (n8n, OpenAI agent builder, Claude SDK).
• Strong Python, with PyTorch / TensorFlow / Hugging Face Transformers.

Responsibilities

• Own features end-to-end from design to UI.
• Design the agentic pipeline and backend orchestration.
• Build the React interface the customer touches.
• Own evals and observability to QA AI output.

Skills

Python

PyTorch

TensorFlow

Transformers

LLMs

Agent frameworks

AI/ML

Eval pipelines

Tools

n8n

OpenAI agent builder

Claude SDK

Job description

$150K–$300K base, and they'll go past $300K for someone exceptional...

As little as two years' experience.

And you have to be in Palo Alto five days a week.

That last bit will lose most of you.

But if you're still reading, this is one of the better AI product roles I've seen this year: Seed-stage LegalTech, ~70 people, $20M in funding. They're rebuilding US immigration - seasoned attorneys plus ML, on a process that decides where real people get to live. The CTO spent five years building the ML platform at one of the big consumer tech companies before this, so it's a proper engineering shop rather than a wrapper with a waitlist.

The role is Product Engineer (AI), and they mean both halves of it.

You'd own features end to end. Design the agentic pipeline, build the backend that orchestrates it, then build the React interface the customer actually touches. No throwing an API over the fence for someone else to integrate later.

You'd also own evals and observability. Deciding how anyone knows the AI output is any good, and catching the regression before the customer does. That's the bit most people skip, and it's the bit they care most about.

What they're actually screening for:

• 2+ years in engineering with AI/ML, LLMs and evaluations. Genuinely open to young engineers who've shipped.
• Agentic workflows built and evaluated in production, across backend and frontend. Research prototypes don't count here - they've already rejected someone on it.
• Real hands-on with agent frameworks; n8n, OpenAI's agent builder, Claude Agent SDK or equivalent. They'll ask you to describe your experience with each, so a skills-list mention won't survive the screener.
• Strong Python, with PyTorch / TensorFlow / Hugging Face Transformers.
• You've come from a company with a genuinely complex, technical product.
• AI-native in how you code. Coding agents as a multiplier, not a curiosity.

Not for you if you want an ML research seat. They've been clear it's a product engineer moving into AI, not the other way round.

Two more honest bits.

It's onsite in Palo Alto, five days, and it's a seed-stage startup that says out loud it's often a 50–60 hour week. They ask about the pace at screening stage, so at least nobody's discovering it in month three.