[BUG] Summary panel hidden in normal layout and Ollama/Qwen reasoning text appears in generated summary

Bug Description

I found two related issues while testing Meetily v0.4.0 on Windows.

First, after generating an AI summary, the summary may not be visible in the normal desktop layout. The app says the summary is ready, but the summary panel does not appear unless I resize the window into a narrow vertical layout. When I do that, the summary becomes visible, but the transcript view disappears.

Second, when using an Ollama model such as qwen3.5:9b, the generated summary may include the model's reasoning/debug text, such as Thinking, Self-Correction, Decision Strategy, and Strict Interpretation. This makes the copied/exported meeting summary unusable as clean meeting notes.

I also noticed that the Enhance feature can sometimes preserve or introduce transcript errors, especially with proper nouns, technical terms, and mixed English/Chinese content.

Current Behavior

1. Meetily generates a summary and shows that the AI summary is ready, but the summary panel is not visible in the normal desktop layout.
2. If I resize the app window into a narrow vertical layout, the summary appears, but the transcript view disappears.
3. When using Ollama with qwen3.5:9b, the generated/copied summary can include reasoning/debug text such as:
   • Thinking
   • Self-Correction
   • Decision Strategy
   • Strict Interpretation
   • Wait...
   • Actually checking the provided text again carefully...
4. The Enhance feature can preserve or introduce incorrect transcript text.

Examples of transcript issues after enhancement:

Intended text	Enhanced transcript output
abnormal outliers	normal outliers
different indicators	different指標
Adi	Eddy
May data	main data
high score	high square

Expected Behavior

1. After a summary is generated, the summary panel should always be visible or accessible in the normal desktop layout.
2. Resizing the window should not make one of the main views disappear without a clear tab or toggle.
3. The transcript and summary should be accessible through a stable layout, such as side-by-side panels or clear Transcript / Summary tabs.
4. When using Ollama models, Meetily should display only the final summary, not the model's reasoning/thinking/debug output.
5. If a model returns reasoning text, Meetily should filter it out before rendering, saving, copying, or exporting the summary.
6. The Enhance feature should avoid changing proper nouns, technical terms, or mixed-language transcript text unless confidence is high.

Steps to Reproduce

1. Open Meetily v0.4.0 on Windows.
2. Use Local Whisper for transcription.
3. Record or import a short meeting transcript.
4. Generate a meeting transcript successfully.
5. Go to AI Model settings.
6. Select Ollama as the summarization provider.
7. Select qwen3.5:9b as the summarization model.
8. Click Generate Summary.
9. Observe that Meetily may show the AI summary is ready, but the summary panel is not visible in the normal desktop window layout.
10. Resize the Meetily app window into a narrow vertical layout.
11. Observe that the summary may appear, but the transcript view disappears.
12. Copy or export the generated summary.
13. Observe that the output may include reasoning/debug text such as Self-Correction, Decision Strategy, or Strict Interpretation.
14. Click Enhance on the transcript.
15. Observe that some transcript errors may remain or become worse.

Environment

• OS: Windows 11
• Browser: Not applicable; Meetily Desktop app
• Version: Meetily v0.4.0
• Node Version: Not sure / not applicable for desktop user
• npm/pnpm Version: Not sure / not applicable for desktop user
• GPU: NVIDIA GeForce RTX 5080 Laptop GPU, 16GB VRAM
• RAM: 32GB
• Transcription model tested: Local Whisper
• Built-in summarization model tested: Qwen 3.5 4B
• Ollama summarization model tested: qwen3.5:9b
• Ollama: running locally

Screenshots/Videos

Should be attached

Error Messages

When using Ollama summarization, I also previously saw this error:

Failed to generate summary: LLM request timed out after 60 seconds

After retrying, the summary could be generated, but the output included reasoning/debug text.

Additional Context

The built-in Qwen 3.5 4B model generated a usable summary.

However, with Ollama qwen3.5:9b, the final copied/generated summary included reasoning-style content. Example phrases included:

• Self-Correction
• Decision Strategy
• Strict Interpretation
• Wait...
• Actually checking the provided text again carefully...

This is problematic because the final meeting summary should contain only clean meeting notes, decisions, and action items.

The test transcript was a short pseudo machine learning project meeting. It included discussion of data cleaning, evaluation metrics, baseline models, GPU experiments, action items, and risks around validation/test set separation.

Possible Solution

Possible fixes:

1. Add a persistent Transcript / Summary tab system so both views are always accessible.
2. Add a layout fallback so the summary panel does not disappear at common desktop window sizes.

Checklist

• I have searched for similar issues
• I have provided all required information
• I have included screenshots/videos if applicable
• I have included error messages if applicable