AI Engineer

About Mirai

At Mirai, we know that a great engineer is defined by their problem-solving DNA, not just the language they wrote in yesterday. Whether you’re a fan of Rust, Python, Java, or Node, we want the architects, the tinkerers, and the lifelong learners.

We are scouting high-potential talent for our exclusive pool, connecting you directly to top-tier startups and global tech leaders.

🚀 Role for our Client Role

You'll be our second AI engineer, working alongside AI Lead in a seven-person engineering team led by the CTO. The team is all internal, full-stack leaning, and ships weekly. We're remote-first with an office in Milan.

TextYess runs AI agents on WhatsApp, onsite chat, and voice (with email launching next). Today, these agents are mostly reactive: they answer customer questions using a RAG pipeline over product catalogs, brand knowledge, and Q&A pairs. The next step is making them genuinely autonomous: agents that take multi-step actions, reason over long conversations, proactively drive revenue, and work across channels with a shared intelligence layer.

That's what you're here to build.

Our agents have handled over 3 million unique customer conversations across 250+ active merchants, with 12 million+ messages sent. The data is there. The challenge now is making the intelligence match the scale.

What you'll work on:

→ Own and improve the RAG pipeline: chunking, re-ranking, context assembly, embeddings

→ Build evaluation frameworks to measure agent performance systematically, not anecdotally

→ Design and ship the multi-agent architecture for cross-channel orchestration

→ Fine-tune models on our proprietary dataset: millions of real merchant-customer conversations, orders, and product catalogs

→ Define the AI capabilities that are uniquely ours

We're an AI-native team: coding agents, agentic workflows, and automated review are part of how we work every day. We expect you to bring that same fluency.

🎯 Who You Are

You've shipped AI systems that real users depend on. You know that evals matter more than vibes, that RAG breaks in production in ways it never does in demos, and that the gap between a good agent and a great one is mostly invisible until you measure it. You have opinions, you push back when something is wrong, and you move fast.

Must-haves:

→ Production experience with LLMs, agents, and multi-agent systems

→ Strong RAG experience: you know when it works, when it doesn't, and why

→ Python skills strong enough to build from scratch

→ Fluent with the modern LLM stack: OpenAI, Anthropic, open-source models

Bonus:

→ Rigorous about evaluation: you don't ship AI without measuring it

→ Experience fine-tuning LLMs for specific domains

→ Cloud infrastructure experience for AI/ML at scale

→ Background in eCommerce or conversational AI

→ Vector databases and semantic search at scale