Lab Assistance Agent Pilot
AI Agent Framework for Laboratory QMS
AI QMS Support
Created an agentic framework to assist lab staff with QMS tasks.
Hybrid Architecture
Evaluated cloud and on-site inference using Jetson Nano.
Slack Integration
Enabled always-on interactions directly inside laboratory Slack channels.
About
Overview
Laboratory environments often require sterile conditions and strict compliance, making hands-free or low-touch interaction highly valuable. In 2024, HEPIUS developed an AI-powered assistance framework designed to support technicians with Quality Management (QMS) tasks, document handling, and internal information retrieval. The pilot focused on integrating multi-agent capabilities into an approachable, chat-based workflow.
Scope of the Project
- Exploration of AI-assisted QMS processes
- Multi-agent workflow design using AWS Bedrock
- Slack-based operational interface tailored for laboratory staff
- Early assessment of offline and on-site inference via Jetson Nano
- Evaluation of RAG and knowledge base accuracy under real conditions
HEPIUS Responsibilities
- Designing the agentic system architecture
- Integrating Slack interactions for seamless user experience
- Assessing LLM performance, hallucination risks, and compliance boundaries
- Evaluating hybrid cloud + edge setups to improve reliability
- Guiding stakeholders through technical and operational decision-making
Technology Components
- AWS Bedrock Agents orchestrating QMS-related tasks
- Serverless backend implemented through AWS Lambda, S3, and DynamoDB
- Slack API integration for mobile-friendly operation
- NVIDIA Jetson Nano evaluation for on-site inference trials
- Anthropic LLM usage for structured responses and document reasoning
Results
AI QMS Support
- Developed an agentic assistant capable of guiding technicians through QMS tasks.
- Enabled structured document support and internal knowledge queries.
- Demonstrated strong usability in sterile laboratory environments.
Hybrid Architecture
- Successfully tested a hybrid AI approach using cloud LLMs and Jetson Nano edge inference.
- Identified scenarios where on-site execution increased responsiveness and privacy.
- Helped stakeholders understand long-term feasibility and infrastructure needs.
Slack Integration
- Delivered a smooth Slack-first interface, aligning with existing laboratory workflows.
- Enabled always-on access for technicians without new tools or UI training.
- Demonstrated strong adoption potential thanks to mobile-friendly interaction.
Pilot Outcome
- Validated technical feasibility but also exposed accuracy limits of 2024-era RAG and LLM models.
- Recommended revisiting production rollout in 2026+ once model reliability improves.
- Stakeholders appreciated the transparency and actionable roadmap for future implementation.
