AI-Assisted Features

QualCoder v2 includes AI-powered features to accelerate your qualitative analysis while keeping you in control.

Info: Human-in-the-Loop

All AI features are designed as suggestions. You always review and approve before changes are made to your data.

Code Suggestions

The AI analyzes your documents and suggests new codes based on patterns it detects.

Getting Suggestions

1. Click AI > Suggest Codes from the menu
2. The AI analyzes your document content
3. Review the suggestions (each shows name, description, and confidence level)
4. Approve to add, Reject to dismiss, or Edit to modify before approving

Duplicate Detection

Over time, you may create similar or redundant codes. The duplicate detector identifies potential matches.

Finding Duplicates

1. Click AI > Find Duplicates
2. The AI compares all codes using token-level similarity (word matching, not character matching)
3. Review candidate pairs (each shows code names, similarity %, and segment counts)
4. Merge A → B to combine codes, or Dismiss if they're not duplicates

Tip: Similarity Threshold - 90%+ - Very likely duplicates - 70-90% - Possibly related, review carefully - Below 70% - Probably distinct concepts

How It Works

Duplicate detection uses word-level (token) matching rather than character-level comparison. This means codes like "Sports & Recreation" and "Trust & Verification" are correctly identified as distinct, even though they share similar character patterns. When both codes have memos, their descriptions are also compared for a more accurate score.

Auto-Code

Automatically apply codes to text matching a pattern.

Pattern-Based Auto-Coding

1. Click AI > Auto-Code
2. Enter a search pattern:
3. Plain text: "interview"
4. Regex: "participant\s+\d+"
5. Select the code to apply
6. Click Preview to see matches
7. Click Apply All to code all matches

The Auto-Code dialog showing pattern entry and match options.

Preview of matches before applying the code.

Auto-Code by Speaker

For transcripts with speaker labels:

1. Click AI > Auto-Code by Speaker
2. Select a speaker (e.g., "Interviewer", "P01")
3. Select the code to apply
4. All text by that speaker is coded

Find Similar

Find passages similar to a coded segment.

Using Find Similar

1. Select a coded segment
2. Click AI > Find Similar
3. Review passages with similar meaning
4. Apply the same code to matches

Tip: Semantic Search

This uses semantic similarity, not just keyword matching. It finds passages with similar meaning even if they use different words.

AI Agent Integration

QualCoder v2 supports AI agents (like Claude Code) working alongside human researchers via the MCP protocol. The agent can:

• Open and close projects programmatically for automated workflows
• Add text sources directly from agent-generated or agent-collected content
• Organize sources into folders (create, rename, delete folders; move sources)
• Remove sources with a safe preview-then-confirm workflow
• Read and analyze document content, suggest codes, and apply coding
• Manage codes — rename codes, update memos, merge overlapping codes, delete irrelevant codes
• Organize themes — create categories, move codes into categories, list category hierarchy

Trust Levels

Agent tools operate at different trust levels for safety:

Level	Meaning	Example Tools
T1 (Autonomous)	Agent acts freely	get_project_context, list_sources
T2 (Notify)	Agent acts, researcher is notified	open_project, close_project
T3 (Suggest)	Agent proposes, researcher confirms	add_text_source, remove_source

Destructive operations like remove_source default to preview mode — the agent shows what would be affected before you approve the action.

Code Management Tools

Agents can perform the full qualitative analysis workflow using these code management tools:

Tool	Purpose	Thematic Analysis Phase
rename_code	Rename a code	Phase 5: Defining Themes
update_code_memo	Set/update code definitions	Phase 5: Defining Themes
create_category	Create theme categories	Phase 3: Searching for Themes
move_code_to_category	Organize codes under themes	Phase 3: Searching for Themes
merge_codes	Consolidate overlapping codes	Phase 4: Reviewing Themes
delete_code	Remove irrelevant codes	Phase 4: Reviewing Themes
list_categories	View thematic structure	All phases

These tools delegate to domain command handlers, ensuring proper event publishing and UI refresh via SignalBridge.

See MCP Setup Guide for configuration and the full list of available tools.

Best Practices

Review All Suggestions - Never blindly accept AI suggestions - Check if suggested codes fit your conceptual framework - Consider if distinctions are meaningful for your research

Iterative Refinement - Start with AI suggestions as a first pass - Refine and merge codes as understanding develops - Use duplicate detection periodically

Document Decisions - Add memos explaining why you accepted/rejected suggestions - Note merge decisions in code memos - Keep an audit trail of AI-assisted changes