Building Specialized AI Assistants with Claude's Projects & Analysis

(Claude access will be provided for workshop activities)

Discover the power of specialized AI agents. This workshop focuses on the "Projects" feature and the native "Analysis" tool. This interactive workshop is about learning Claude internal skills and memory. It answers the beginner's frustration: "Why do I have to keep explaining my business to An AI every time I start a new chat?"

Stop treating Claude like a stranger. In this workshop, you will have an opportunity to learn how to use Claude Projects to build specialized AI assistants that "know" you and your specific needs. We will move beyond simple chats and build persistent "Skills", custom environments where Claude has access to your specific brand guidelines, codebases, or data. Additionally, we will explore the Analysis Tool (Claude's built-in coding sandbox), teaching you how to use Claude for accurate data science and math, eliminating the "hallucinations" common in standard chatbots.

By the end of the workshop, participants will have had an opportunity to build three distinct "Specialists": a Writing Coach, a Data Analyst, and a Project Manager.

Topic Outline

Session 1
Module 1: The Memory Problem
The Issue: Why standard chats are inefficient (The "Goldfish Memory" problem).
The Solution: Introduction to Claude Projects.
Structure: Understanding the difference between "Project Knowledge" (Background Data) and "Custom Instructions" (Behavior).
Activity: Creating your first Project shell.

Module 2: The "Brand Specialist" (Knowledge Management)
Objective: Teaching Claude your voice.
The Knowledge Base: Uploading PDFs, Style Guides, and Past Examples into the Project.
Persistent Knowledge: Explaining how Claude "reads" this whole library instantly before every answer.
Testing: Prompting "Write a LinkedIn post about this workshop." Claude automatically uses your tone without being asked.

Session 2

Module 3: The "Analysis Skill" (Data Science)
Objective: Perfect Math and Logic.
The Feature: Enabling the Analysis Tool.
The Trap: Why LLMs are bad at math (predicting text vs. calculating).
The Fix: Uploading a CSV file and instructing Claude to write Python code to analyze the data.
Activity: "Analyze this sales spreadsheet and generate a visualization of monthly trends."

Module 4: Deep Dive – Complex Analysis
Objective: Expanding data analysis skills.
Combining Sources: Demonstration of uploading multiple files and cross-referencing them (instructor-led).
Visualization Options: Overview of available chart types and when to use them.
Refining Outputs: Participants enhance their Module 3 visualization with one improvement (different chart type, adjusted colors, or added labels).
 

Session 3

Module 5: The "Project Manager" (Workflow Automation)
Objective: A task management assistant that knows specific processes.
Custom Instructions: Configuring a Project specifically for project management.
Defining Workflows: "When a new initiative is described, break it into phases with milestones and dependencies."
Template Library: Uploading team templates, SOPs, and past project plans.
Activity: "Create a launch plan for this product based on a standard process."

Module 6: Advanced Custom Instructions (Prompt Engineering)
Objective: Locking in the behavior.
Role Definition: "You are a Senior Editor. You are cynical and strict. You hate adverbs."
Output Formatting: Forcing Claude to always answer in a specific table format or JSON.
Refining: Tweaking the "System Prompt" of the Project until the behavior is perfect.
Objective: A coding partner that knows your stack.
Custom Instructions: Configuring a Project specifically for code.
Defining Constraints: "Always use Python 3.10. Never use 'pandas' unless necessary. Always comment your code."
Snippet Library: Uploading your team's API documentation so Claude stops inventing fake APIs.

Session 4

Module 7: The "Research Assistant" (Web Search & Synthesis)
Objective: Combining web search with Projects for research workflows.
The Feature: Enabling web search within a Project environment.
New Skill: Teaching Claude to search, evaluate sources, and synthesize findings.
Custom Instructions: "When researching a topic, always cite sources and flag conflicting information."
Activity: "Research the top three competitors in [industry] and summarize their pricing strategies."
 

Module 8: Troubleshooting & Best Practices
Objective: Solving common problems and planning for the future.
Updating Knowledge: When to delete old files from the Project to keep Claude "fresh."
The "Context Window" Limit: How to manage project size to avoid running out of memory.
When Claude Ignores Instructions: Diagnosing and Fixing Instruction Conflicts.
When Outputs Drift: How to tighten prompts for consistency.
When Analysis Fails: What to do when code execution produces unexpected results.
Identifying New Specialists: How to spot tasks that benefit from a specialist (repetitive, context-heavy, format-specific).
Next Steps: Resources for continued learning and experimentation.

Q&A and Showcase
Prerequisite Checklist
This workshop relies heavily on the Claude Projects feature, which is gate-kept behind the paid subscription.

Required Hardware
Mac or Windows computer with Chrome or Edge browser.