Agentic AI Software Developer

Job Description

Responsibilities:

• Design, build, and evaluate agentic AI systems that can plan, reason, act, and collaborate across tools and environments, including single- and multi-agent setups for complex, long-horizon tasks.
• Develop robust agent harnesses and evaluation frameworks covering end-to-end testing, regression analysis, trace logging, replayability, and metrics for success, cost, latency, robustness, and safety.
• Implement self-improving agent loops using reflection, critique, self-debugging, and iterative optimization strategies driven by agent experience, execution traces, and automated feedback.
• Architect and optimize agent memory systems, including short-term and long-term memory, retrieval-augmented generation, summarization, compression, forgetting policies, and privacy-aware retention.
• Enable reliable deployment of agents on constrained and edge environments, focusing on model/runtime optimization, partial or offline execution, secure tool-use, and seamless edge-cloud coordination.

Qualifications

Basic Qualifications

• Bachelor or master's degree in computer science or engineering
• Strong software engineering skills in Python
• Hands-on experience with LLMs and agentic systems, such as tool-using agents, planner-executor patterns, multistep reasoning pipelines, RAG systems
• Solid understanding of ML fundamentals and practical model usage, including prompting, evaluation, and error analysis
• Ability to design experiments and interpret results, including ablations, statistical thinking, clear success criteria and measurable KPIs
• Strong communication and documentation skills including clear write-ups, reproducible experiments, and crisp technical presentations

Preferred Qualifications

• 3+ years experiences in industrial research.
• Experience with one or more of the following topics:
• Agent frameworks
• Structured generation
• Agent memory management
• Self-improving agents
• LLM fine-tuning & reinforcement learning
• Model optimization for edge