Machine Learning Engineer

Founding Machine Learning Engineer – Transformers, RAG, and Agentic Systems

We're working with a stealth startup backed by one of Silicon Valley's top investors, building foundational infrastructure for AI systems that reason, retrieve, and interact — not just generate.

This is an opportunity to join as a founding engineer focused on transformer-based architectures, evaluation pipelines, and LLM-native agents. You'll be part of a small, senior team solving hard problems around performance, grounding, and generalization in LLM-based systems.

What You'll Be Working On:

Fine-tuning and optimizing transformer models for real-world task performance (instruction tuning, LoRA, QLoRA)

Designing and deploying RAG pipelines using vector stores, custom embedding strategies, and multi-hop retrieval

Architecting agentic systems that can reason, plan, and use tools through language interfaces (LangChain/LangGraph-style frameworks)

Developing robust evaluation frameworks to measure hallucination, factuality, retrieval relevance, and tool use accuracy

Working with deeply technical founders on building from first principles — no legacy baggage, no corporate layers

You Might Be a Fit If You:

Have shipped or trained LLMs, or have gone deep into adapting/optimizing them (e.g. LLaMA, Falcon, Mixtral, Phi)

Are fluent with modern LLM tooling – HuggingFace Transformers, DeepSpeed, vLLM, PyTorch Lightning

Understand how to build and evaluate retrieval-augmented generation (RAG) systems using FAISS, Chroma, or Pinecone

Have built or contributed to tool-using agents or function-calling workflows, and know the edge cases

Are comfortable building in a lean, highly technical environment with direct ownership and high ambiguity

Why This Role?

You'll be building the core of something truly ambitious — not another LLM wrapper, but real systems intelligence

You'll work with a team that cares deeply about technical rigor, practical grounding, and speed of execution

And you'll have the chance to define not just the system, but the standard for how LLMs are evaluated, used, and trusted