Responsibilities
- Lead the architecture and ownership of robust multi-agent systems using frameworks such as LangChain, LangGraph, CrewAI, OpenAI Agents SDK, AutoGen/AG2, and Semantic Kernel, with careful consideration of state, routing, memory, and error handling.
- Develop cognitive architectures for agents, including planning mechanisms like ReAct, Reflexion, and Chain-of-Thought, tool interaction models, memory structures, and self-assessment loops.
- Create patterns for multi-agent coordination, such as supervisor-worker hierarchies and peer collaboration, aligned with open standards and integrated via MCP servers into enterprise systems, clinical databases, and regulatory sources.
- Design and manage shared AI infrastructure, including LLM routing, embedding services, vector databases, RAG workflows, prompt management, and model evaluation tools across all agent-based products.
- Define model governance strategies covering both cloud-hosted models (e.g., Claude, GPT, Gemini) and self-hosted variants, including fine-tuning pipelines using LoRA/QLoRA for drug development use cases.
- Establish context engineering practices for managing retrieval, chunking, re-ranking, hybrid search, and query routing across large-scale clinical and scientific datasets, incorporating safety filters and human-in-the-loop escalation suitable for GxP settings.
- Define the full software development lifecycle for agent systems, treating agent behavior as a controlled software component subject to versioning and change management.
- Develop evaluation frameworks using test sets, LLM-based scoring, regression tracking, task completion metrics, and performance dashboards, integrated into CI/CD pipelines with automated gates, drift detection, and rollback features.
- Ensure traceability, audit logging, and version control to support compliance with GxP, 21 CFR Part 11, and internal AI governance policies.
- Implement comprehensive observability across the stack using tools like LangSmith, Langfuse, and OpenTelemetry, including trace logging, token usage tracking, latency analysis, and anomaly detection.
- Ensure system reliability through retry mechanisms, fallbacks, circuit breakers, and graceful degradation, while monitoring for behavioral drift and decision inconsistencies using feedback loops that avoid regressions.
- Integrate agent services with enterprise systems such as Salesforce, MuleSoft, Veeva, SAP, Databricks, and ServiceNow using MCP and standardized API designs.
- Design authorization models based on AI risk classifications, specifying agent access, action rights, and autonomous decision boundaries versus human oversight requirements.
- Enforce governance aligned with FDA AI/ML guidance, ICH E6/E8, EU AI Act, and internal policies, ensuring data residency, privacy (HIPAA, GDPR), least privilege access, prompt injection protection, and secure integrations.
- Produce validation documentation to meet audit requirements for agents used in clinical, regulatory, and GxP-controlled processes.
- Collaborate with product managers, data scientists, architects, security teams, and domain experts to convert pharmaceutical challenges into scalable agent system designs and reusable platform components.
- Mentor engineering teams, lead architecture reviews, set technical standards, and promote production-grade AI development while advancing adoption of emerging protocols like MCP, A2A, and evaluation benchmarks.
Work Arrangement
Hybrid
