Scientific Agent / Autonomous Research Scientist & Engineer

Scientific Agent / Autonomous Research Scientist & Engineer

Role Overview

You will work on the core R&D of Scinetics Scientific Agent and Auto-Research systems, enabling AI to start from open-ended scientific questions and autonomously decompose problems, retrieve literature and data, select and use tools, execute code and scientific computation, analyze results, reflect, and iterate. Our goal is to move toward AI systems capable of long-horizon scientific research and autonomous discovery.

This is not a simple RAG, chatbot, or workflow automation role. We care about how agents can work continuously on real scientific tasks, use complex research tools correctly, adapt their research path based on intermediate results, and ultimately produce scientifically verifiable outcomes.

Research Areas

• Scientific Agent / Autonomous Research

• Deep Research / Auto-Research

• Long-horizon Agent & Agentic Reasoning

• Tool Use & Tool Creation

• Multi-Agent Systems

• Agent Memory & Context Management

• Scientific Coding & Computational Research

• Agent Evaluation / Verifier / Reward

• Self-improving Agents

• Computer Use / GUI Agent

• AI for Science

Responsibilities

• Scientific Agent: Design and build Scientific Agent / Auto-Research systems covering the full research loop from scientific problem understanding and task planning to information retrieval, tool use, code execution, result analysis, and self-correction.

• Long-horizon Research: Explore long-horizon planning, agent memory, context management, reflection, tool discovery / creation, and multi-agent collaboration to improve autonomy and stability on complex scientific tasks.

• Scientific Tool Use: Connect agents to real scientific tools and environments, including scientific databases, Python / R, bioinformatics, molecular modeling, simulation, scientific software, and Web / GUI interfaces, continuously expanding the set of research tasks agents can complete autonomously.

• Agent Evaluation: Build scientific-agent evaluation systems using real research tasks, trajectory quality, tool-use success, scientific correctness, and end-to-end task completion to drive iteration.

• Self-improvement: Work with foundation-model, post-training, and scientific-data teams to explore the continuous learning loop of agent trajectory → training data → model improvement → better agent.

What We Look For

• Bachelor’s degree or above in computer science, artificial intelligence, mathematics, or a related STEM field. Exceptional interdisciplinary candidates are welcome regardless of major.

• Hands-on R&D experience with LLM agents, Auto-Research, Deep Research, coding agents, computer use, or other autonomous-agent systems.

• Strong Python engineering skills and the ability to independently take a system from prototype to deployment, evaluation, and iteration.

• Familiarity with tool use, function calling, agentic workflows, sandboxing, and code execution.

• Understanding of common agent failure modes, including hallucination, tool failure, long-context degradation, error propagation, and trajectory-quality issues.

• Ability to show real agent projects, outputs, demos, trajectories, papers, or open-source work.

• Long-term interest in AI for Science and autonomous scientific discovery.

Nice to Have

• Experience building research or coding agents that can operate autonomously for extended periods.

• Experience with Scientific Agents, AI4S agents, or autonomous research systems.

• Experience with agent benchmarks, evaluation, verifiers, RL, or RLVR.

• Experience with tool creation, MCP, sandbox execution, or related infrastructure.

• Experience with GUI agents / computer use.

• Research or engineering experience in multi-agent systems, agent memory, or long-context systems.

• Publications at strong conferences or journals.

• High-quality agent open-source projects or GitHub contributions.

If you can show us a real trajectory or demo in which an agent receives a complex task, plans autonomously, uses tools, executes, iterates, and completes the task, that will tell us far more than a list of agent frameworks on your resume.