Product ML Engineer

Alpaca Research; Cambridge (MA); On-site; USD 160,000 - 180,000; Full time

Job summary

Alpaca Research is hiring an ML engineer to bridge the gap between raw model capabilities and full system integration. You will work on model optimization, agent architectures, memory, and evaluation to build capable, efficient, and reliable local AI products.

You’ll experiment with fine-tuning, memory management, and inference-time techniques, translating research ideas into production systems used by real users.

Qualifications

• Strong ML fundamentals with modern language models, transformers, PyTorch, and training/inference techniques.
• Experience with LLM agents, tool calling, planning, retrieval, or multi-step reasoning systems.
• Understanding of long-context processing and memory architectures for context windows.
• Familiarity with quantization, efficient inference, and tradeoffs between quality, latency, and compute.
• Experience with post-training techniques (supervised fine-tuning, RL, distillation).
• Interest in reading papers and turning research into working software.
• Ability to own ambiguous problems and thrive in experiments.

Responsibilities

• Fine-tune, distill, and optimize language models for tool use, instruction following, and complex agentic tasks.
• Develop harnesses, tool-use architectures, planning/execution loops to improve model capabilities and reliability.
• Build retrieval and memory systems for long tasks and conversations.
• Experiment with inference-time techniques and local hardware optimization.
• Develop benchmarks and evaluations for agent capabilities and real-world performance.
• Prototype and implement promising ideas from research into production.

Skills

ML fundamentals

Agentic systems

Memory & context

Model optimization

Post-training

Research mindset

Independence

PyTorch

Transformers

Tools

PyTorch

Transformers

Hardware/ML systems

Job description

Compensation: $160,000 - $180,000 + Equity

Who we are

Today, AI is something you rent. Alpaca is building the full stack of hardware, software, models, and applications together to make it something you own.

We are developing a new computer architecture that breaks the economics keeping frontier models trapped in datacenters. On top of that, we’re building a platform of models, applications, and services that give that hardware meaning.

Personal AI is moving from the datacenter to the desk, and we are driving that change.

Our team is based in Cambridge, MA.

About the role

We want to make local AI models capable of doing far more than answering questions. Our goal is to build intelligent systems that can understand their users, remember information over time, use tools, and reliably accomplish complex tasks on their behalf.

We are looking for an ML engineer to help bridge the gap between what a model can do on its own and what it can accomplish as part of a complete system. You'll work across model optimization, post-training, agent architectures, memory systems, and evaluation to make our AI more capable, efficient, and reliable.

This is a role for someone who enjoys both ML research and building real systems. You should be comfortable fine-tuning models, experimenting with inference-time techniques, designing agent harnesses, and developing new approaches to managing context and memory. You'll have the opportunity to experiment with ideas, measure what works, and bring improvements directly into products used by real people.

What you’ll do

• Improve our models: Fine-tune, distill, and optimize language models for tool use, instruction following, and complex agentic tasks.
• Build agent systems: Develop harnesses, tool-use architectures, planning and execution loops, and other techniques that make models more capable and reliable.
• Engineer memory and context: Build retrieval, persistent memory, context compression, and management systems that allow agents to work effectively across long tasks and conversations.
• Optimize inference: Experiment with quantization, speculative decoding, and inference-time techniques to improve model quality, speed, and efficiency on local hardware.
• Evaluate and iterate: Develop benchmarks and evaluations for agent capabilities, long-horizon reliability, memory, and real-world task performance.
• Research and implement: Explore new papers, algorithms, and techniques, rapidly prototype promising ideas, and integrate successful approaches into production.

Who we’re looking for

• Strong ML fundamentals: Experience with modern language models, transformers, PyTorch, and the training and inference techniques behind them.
• Agentic systems experience: You've built or experimented with LLM agents, tool calling, planning, retrieval, or multi-step reasoning systems.
• Memory and context engineering: Understanding of long-context processing, memory architectures, and techniques for working within limited context windows.
• Model optimization: Familiarity with quantization, efficient inference, and the tradeoffs between model quality, latency, and compute.
• Experience with post-training: Familiarity with supervised fine-tuning, reinforcement learning, distillation, or other techniques for improving model behavior.
• Research-minded engineering: You enjoy reading papers, implementing new ideas, and turning research into working software.
• Independence: You take ownership of ambiguous problems, move quickly, and thrive in an experimental environment.