Agentic AI Engineer

About Elemynt

Elemynt, an early-stage venture by Xora Innovation, specializes in integrating AI into practical applications. The platform merges sophisticated machine learning, high-performance simulations, and modern software engineering to speed up the creation, validation, and deployment of new materials, operating at the intersection of AI, physics, and extensive computational methods. The challenges are complex, with significant real-world impact.

Role Overview

This position involves crafting the foundational AI agent layer that supports all large language model (LLM)-enabled features on the platform. Responsibilities include creating an interface for multiple model providers, orchestrating multi-step agent workflows, managing retrieval and prompt systems, and implementing tracing and evaluation mechanisms to ensure reliability and transparency.

Key Responsibilities

• Develop provider abstractions allowing workflows to flexibly call, switch, or add model providers via configuration, handling commercial APIs and self-hosted endpoints with structured output validation, retry logic, and cost monitoring.
• Design and implement agent orchestration that uses a planning agent to distribute tasks to specialized sub-agents in parallel within a stateful framework, including durable checkpoints, conditional branching, and memory/context management for coherent multi-step workflows over extended tasks.
• Integrate human-in-the-loop checkpoints for low-confidence or high-risk operations to ensure human oversight before proceeding.
• Adapt existing platform functionalities into typed and registered tools invoked by agents, maintaining clear separation between the agent layer and underlying systems.
• Build retrieval pipelines end-to-end — from data ingestion, embedding generation, chunking, through hybrid search and reranking — assembling coherent context grounding each model call.
• Manage the prompt layer with versioned prompts, few-shot examples, and captured reasoning records to ensure traceability and inspectability of every call.
• Provide a unified integration point exposing agents and safeguarded model access tools for backend services, frontends, and notebooks.
• Instrument every model invocation, tool usage, and agent execution as traced spans annotated with prompt, model, and tool lineage to facilitate debugging and cost control.
• Create an evaluation framework combining deterministic trace metrics and LLM-as-judge scoring to ensure faithfulness, enabling pre-release gating and regression detection.

Qualifications

• Bachelor’s or Master’s in Computer Science or related engineering discipline with over 5 years of experience developing and deploying production software, particularly deep expertise building LLM or agent systems at scale.
• Proficient in Python programming, including asynchronous code, typing, modular design, rigorous testing, and code reviews; a demonstrated record of delivering dependable systems.
• Practical experience in constructing agentic or LLM systems with orchestration loops, tool integrations, structured outputs, and reliable context/memory handling for prolonged workflows.
• Familiarity with managing multiple model providers within a unified abstraction, including routing, fallback strategies, and cost-latency tradeoffs.
• Hands-on experience developing retrieval systems encompassing embedding techniques, chunking, combined search, reranking, and vector database utilization.
• Expertise in LLM evaluation and safety guardrails — developing evaluation datasets, harnesses, LLM-based scoring, regression controls, and output quality/safety validations.
• Skilled in adding observability to LLM systems, including tracing of model and tool calls, prompt versioning, and leveraging traces for debugging and system enhancement.
• Comfortable with full ownership of complex, evolving systems in a dynamic, early-stage startup setting.

Desirable Skills

• Experience with stateful agent orchestration frameworks such as LangGraph or AutoGen, and durable execution engines like Temporal for managing lengthy workflows.
• Background in building MCP tools, servers, or similar tool-calling integration frameworks.
• Familiarity with ML Ops and evaluation platforms such as MLflow or Langfuse for prompt versioning, tracing, and performance evaluation.
• Knowledge of human-in-the-loop and interrupt-driven agent designs to enable controlled review processes.
• Applied experience deploying LLMs within scientific or technical workflows grounded in tool outputs and structured data.
• Engagement with modern AI coding assistants or open-source contributions to AI/agent tooling communities.

Location

Positions are available in Singapore and the United States, with onsite or hybrid work modes determined by location.

Additional Information

Candidates who do not meet every qualification but demonstrate strong interest and aptitude are encouraged to reach out.