Building a Research Assistant Agent

Building a Research Assistant Agent

Create an AI agent that becomes an expert in your specific research domain, trained on your papers, protocols, and institutional knowledge.

Use Case Overview

Problem: Your lab has accumulated years of protocols, papers, and tribal knowledge. New team members spend weeks getting up to speed, and even experienced researchers forget details of rarely-used procedures.

Solution: Build a Research Assistant agent trained on your lab’s documentation. Team members can ask questions in natural language and get accurate, sourced answers instantly.

Step-by-Step Guide

Step 1: Gather Your Knowledge Base

Collect documents that represent your domain expertise:

• Lab protocols (PDF, Word, Markdown)
• Published papers from your group
• Equipment manuals and SOPs
• Meeting notes with methodological decisions
• Troubleshooting guides from past issues

Tip: Start with 5-10 high-quality documents. You can always add more later.

Step 2: Create the Agent

1. Navigate to AI Agents in the sidebar
2. Click Create Agent
3. Configure the basics:

Step 3: Write Effective Instructions

Your agent’s system prompt defines its expertise and behavior. Here’s a template:

You are a protein crystallography expert for the Smith Lab at University X.

Your expertise includes:
- Crystal growth and optimization protocols
- X-ray diffraction data collection
- Structure solution using molecular replacement
- Refinement with Phenix and CCP4
- Our lab's specific equipment (Rigaku MicroMax-007)

When answering questions:
- Reference specific protocols from the knowledge base
- Cite relevant papers when applicable
- Warn about common pitfalls we've documented
- Suggest consulting Dr. Smith for novel situations

You have access to our complete protocol library, equipment manuals,
and published methods papers.

Step 4: Upload Knowledge Base Documents

1. Select the Knowledge Base tab

2. Click Add Resources

3. Choose your source:

   • From Stash: Files already in your workspace
   • From URL: Online documentation
   • From Google Drive: Shared team folders
   • From S3: Cloud storage buckets

4. Wait for indexing to complete (status changes to “Indexed”)

Step 5: Test Your Agent

Open a Nova conversation and invoke your agent:

@myusername.crystal_expert What's our standard protocol
for growing lysozyme crystals? Include the specific
PEG concentrations we use.

The agent will search your knowledge base and provide a sourced answer.

Example Interactions

Protocol Lookup

User: @lab.crystal_expert How do I prepare the cryoprotectant
for flash-freezing?

Agent: Based on Protocol #CR-007 in our knowledge base,
the standard cryoprotectant preparation is:

1. Start with mother liquor from the crystal drop
2. Add glycerol to 20% (v/v) final concentration
3. Mix gently to avoid precipitation
4. Transfer crystal using a nylon loop

Note: For PEG-based conditions, you may need to increase
glycerol to 25% per Dr. Smith's 2023 optimization study.

Troubleshooting

User: @lab.crystal_expert My crystals are growing too fast
and have poor diffraction. What should I try?

Agent: According to our troubleshooting guide (Section 4.2),
rapid crystal growth often indicates supersaturation issues.
Recommended adjustments:

1. Reduce protein concentration by 20-30%
2. Lower precipitant concentration by 5%
3. Try seeding with microcrystals from a previous batch

Our 2022 paper on optimization (Smith et al., Acta Cryst)
found that slower growth at 4C improved resolution
from 3.2A to 2.1A for similar cases.

Best Practices

Document Quality Matters

• Well-structured documents with clear headings work best
• Include context and rationale, not just procedures
• Update documents when protocols change

Keep Instructions Focused

• Define a clear scope for the agent
• Specify when to defer to human experts
• Include domain-specific terminology

Iterate Based on Usage

• Review agent responses periodically
• Add documents that fill knowledge gaps
• Refine instructions based on common questions

Sharing with Your Team

1. Go to your agent’s settings
2. Under Sharing, select your team
3. Team members can now invoke: @teamname.crystal_expert

Related: Knowledge Bases | Agent Instructions | Data Analyst Agent