AI Engineer

• Salary
  : $130k – $170k 0.0% – 0.2%
• Location
  :
  Remote (United States)
• Experience
  : 2 years of exp
• Job Type
  :
  Full Time
• Visa Sponsorship
  :
  Not Available
• Skills
  :
  Python, LLMs

Arre Path is a fully remote startup using AI to discover new classes of oral small-molecule drugs. We work in antibacterials, mycobacterial lung disease, immunology, and inflammation. Our platform has found inhibitors of six new antibacterial targets. Our team has helped discover, develop, or obtain approval for nine antibiotics. Our investors include Insight Partners, Innospark Ventures, and the Boehringer Ingelheim Venture Fund.

Ariadne is our AI scientific assistant. It is an agent with persistent memory, a web chat interface, and a single MCP gateway that brokers access to compute on AWS (Batch, Lambda, S3), external scientific APIs, and our internal data. It also includes a multi-stage literature search pipeline and an evaluation harness for measuring agent behavior. It gives our scientists natural-language access to more than 30 machine-learning models and cheminformatics tools.

They use it to run computational chemistry workflows, search the scientific and patent literature, query internal assay data, and analyze results in conversation multiple times per day.

The role

We're hiring an engineer to be the second developer on Ariadne. You'll work directly with Kurt Thorn, our CTO, who leads Ariadne development, and with the scientists who use it every day. Arre Path is a small company, so you'll get direct exposure to the leadership team.

On a two-person team you own work end to end. That means taking a bug report or feature request from a scientist, designing the change, building it with coding agents, validating it, and shipping it to production. There's no product manager, QA team, or separate ops group. You'll be the person who decides how something should work and makes sure it does.

Your main focus will be making Ariadne more autonomous, reliable, and robust:

• Memory: what the agent retains across sessions, and how it retrieves and uses it
• File tracking: keeping track of the data files, results, and artifacts a scientist works with over a long project
• Reasoning: planning and staying on course through long, multi-step scientific tasks
• Tool execution: choosing the right tools, calling them correctly, and recovering when they fail
• Delegation: handing work off to sub-agents and combining what they return
• Web interface: improving the Type Script and Vite chat app where it limits what scientists can do
• The stack is Python, Type Script, AWS, Docker, Git Hub Actions, and LLMs from several providers via Amazon Bedrock.

Two humans and a fleet of coding agents build Ariadne. Most of our code is written by agents such as Claude Code and Codex, running in parallel. Our job is to direct them well and to be accountable for what ships.

• Humans own design and judgment. Bounded changes get a short design review. Architectural changes get a written spec and an implementation plan before any code is written.
• Tests and evidence over assertions. We practice TDD for behavior and verify agent claims against real results.
• Written in variants. Component specs record the in variants that matter, and tooling checks code against them.
• Cross-model review. We use reviewers from different model families to stress-test designs and diffs.
• Repository rules agents follow. Our CLAUDE.md, skills, and hooks encode how work gets done, and we keep improving them.
• You'll help shape this process, not just follow it. If you've built your own agentic workflow and have opinions about what works, we want to hear them.

The essential qualification is experience on both sides of agentic software:…