Software Engineer - AI Accelerated Development

Signant Health develops technology and services for clinical trials. The Software Engineer – AI Accelerated Development builds AI-powered agentic features for clinical trial platforms, integrating LLMs and delivering production-ready applications with appropriate evaluation, observability, and risk controls.

Responsibilities

• Design and build AI agents and agentic workflows — tool-use/function calling, multi-step task orchestration, and agent loops — that power Signant Health's clinical trial platforms
• Integrate LLM APIs (Anthropic, OpenAI, Bedrock, or Azure OpenAI) into .NET application code, handling authentication, streaming, error handling, and cost/latency tradeoffs
• Deliver AI features end-to-end, taking them from backend integration through to a usable, production-ready front-end
• Build custom agents and tools with the Claude Agent SDK, wiring and steering agent loops
• Author custom skills and slash commands for Claude Code to codify team-specific workflows
• Implement RAG pipelines and integrate vector stores (e.g. Qdrant, pgvector, Pinecone) to ground agent outputs in trusted data
• Instrument agents for evaluation and observability so their outputs can be verified, measured, and audited
• Apply AI-specific risk controls appropriate to a regulated clinical environment, addressing hallucination, determinism, and auditability of agent decisions
• Use Claude Code (or an equivalent agentic CLI) daily as a primary development tool, reviewing and steering AI-generated diffs

Preferred

• comfort with async patterns
• Claude Agent SDK experience — building custom agents, defining tools, and wiring agent loops
• Working knowledge of agent primitives: hooks (deterministic pre/post-tool-call controls), skills (progressive-disclosure, load-on-demand instructions), subagents/multi-agent delegation, and session/state management
• Experience authoring custom skills or slash commands for Claude Code to codify team-specific workflows
• Exposure to MCP (Model Context Protocol) for connecting agents to internal tools and data sources
• Working knowledge of RAG patterns and vector store integration (e.g. Qdrant, pgvector, Pinecone)
• Awareness of AI-specific risk areas relevant to a regulated environment (OWASP LLM Top 10, hallucination/determinism concerns, and auditability of agent decisions)
• Asynchronous Programming
• Claude Agent SDK
• Custom Agent Development
• Tool Definition
• Agent Loop Design
• Agent Hooks
• Progressive Disclosure
• On-Demand Instructions
• Subagent Delegation
• Multi-Agent Orchestration
• Session Management
• State Management
• Claude Code Skills
• Slash Command Development
• Workflow Codification
• Model Context Protocol (MCP)
• RAG
• Vector Store Integration
• Qdrant
• pgvector
• Pinecone
• AI Risk Awareness
• OWASP LLM Top 10
• Hallucination Mitigation
• Deterministic Agent Behavior
• Agent Decision Auditability

Benefits

• Opportunities to learn, take ownership, and drive meaningful innovation.
• A collaborative, global environment where you can grow your career.

Company

• Clinical trial technologies and services for drug research
• Blue Bell, Pennsylvania, USA
• https://www.signanthealth.com/
• Founded in 2000
• 1001-5000 employees

Qualifications

• Full-stack .NET development — C#/ASP.NET Core on the backend paired with a modern front-end framework
• Hands-on agentic development experience — building AI agents or agentic workflows (tool-use/function calling, multi-step task orchestration, agent loops); not just prompting an LLM, but building systems around one
• Practical LLM API integration (Anthropic, OpenAI, Bedrock, or Azure OpenAI), including authentication, streaming, error handling, and cost/latency tradeoffs
• Proven full-stack delivery of AI features — able to take an agentic/AI feature from backend integration through to a usable front-end, not just a notebook prototype
• Daily, hands-on use of Claude Code (or an equivalent agentic CLI) as a primary development tool — comfortable working through multi-step agentic sessions and reviewing and steering AI-generated diffs, rather than occasional autocomplete-style use
• Prompt engineering — designing and iterating on system prompts as production software contracts, not one-off experiments
• Python proficiency as the preferred language for AI/agent tooling
• Evaluation and observability literacy — able to reason about whether an agent's output is correct and instrument it to prove so
• Comfort with async patterns
• Claude Agent SDK experience — building custom agents, defining tools, and wiring agent loops
• Working knowledge of agent primitives: hooks (deterministic pre/post-tool-call controls), skills (progressive-disclosure, load-on-demand instructions), subagents/multi-agent delegation, and session/state management
• Experience authoring custom skills or slash commands for Claude Code to codify team-specific workflows
• Exposure to MCP (Model Context Protocol) for connecting agents to internal tools and data sources
• Working knowledge of RAG patterns and vector store integration (e.g. Qdrant, pgvector, Pinecone)
• Awareness of AI-specific risk areas relevant to a regulated environment (OWASP LLM Top 10, hallucination/determinism concerns, and auditability of agent decisions)

Required skills

• AI Agent Development, AI Evaluation and Observability, ASP.NET Core, C#, Claude Agent SDK, Claude Code, LLM API Integration, Model Context Protocol (MCP), Prompt Engineering, Python, Retrieval-Augmented Generation (RAG), Vector Stores

Preferred skills

• comfort with async patterns

Experience

• 1-3 years