13 - Advanced Workflows Guide

13 - Advanced Workflows Guide

⚡ Power User Features & Optimization Strategies
⏱️ Time Estimate: 20 minutes
📋 What You’ll Learn: Hybrid setups, cost optimization, batch processing, power user techniques

---

Table of Contents

• Hybrid Privacy Setup
• Cost Optimization Strategies
• Batch Processing Workflows
• Job Queue Management
• Keyboard Shortcuts Reference
• Notification Management
• Widget Customization
• Custom Model Integration

---

Hybrid Privacy Setup

Best of Both Worlds

Combine local privacy with cloud performance strategically.

Configuration 1: Private Audio + Cloud Analysis

Setup:

Transcription:  Ollama (whisper:base)
Analysis:       OpenAI (gpt-4o-mini)

Benefits:

• 🔒 Audio never leaves device (max privacy)
• 💰 Save transcription costs (~$0.36/hour)
• ⚡ Fast cloud analysis (~10-20 seconds)
• 📊 High-quality insights

Use cases:

• Confidential meeting audio
• Cost-conscious with high volume
• Privacy regulations (HIPAA, GDPR)

Cost analysis:

1-hour meeting:
- Transcription: Free (local)
- Analysis: ~$0.02-0.05 (cloud)
- Total: ~$0.02-0.05

vs Full Cloud:
- Transcription: $0.36
- Analysis: $0.02-0.05
- Total: $0.38-0.41

Savings: 90%+ on audio transcription

Configuration 2: Cloud Speed + Local Review

Setup:

Transcription:  OpenAI (whisper-1)
Analysis:       Ollama (llama3.1)
Auto-analyze:   Disabled

Benefits:

• ⚡ Fast transcription (~30 seconds)
• 🔒 Analysis stays local (no insights to cloud)
• 🧐 Manual review before processing
• 💰 Free analysis

Use cases:

• Fast transcript turnaround needed
• Sensitive insights (strategies, plans)
• Review before AI processing

Cost analysis:

1-hour meeting:
- Transcription: $0.36 (fast cloud)
- Analysis: Free (local)
- Total: $0.36

Plus: Control over what gets analyzed

Configuration 3: Test Local, Production Cloud

Development:

Transcription:  Ollama (whisper:base)
Analysis:       Ollama (llama3.1)

Production (important meetings):

Transcription:  OpenAI (whisper-1)
Analysis:       OpenAI (gpt-4o)

Benefits:

• 🧪 Free testing and iteration
• ⭐ Best quality for important content
• 💰 Spend only where it matters

Workflow:

1. Test workflow with Ollama (free)
2. Verify prompt quality
3. Switch to cloud for final version
4. Get best results

---

Cost Optimization Strategies

Strategy 1: Selective Cloud Usage

Principle: Use local for routine, cloud for critical

Implementation:

Create two project categories:

📁 Routine (Local Processing)
   └─ Daily standups
   └─ Internal team syncs
   └─ Voice notes

📁 Critical (Cloud Processing)
   └─ Executive meetings
   └─ Client presentations
   └─ Strategic planning

Savings example:

Routine (20 hours/month):
- Local: Free
- Cloud: $15.20

Critical (5 hours/month):
- Cloud: $3.80

Total monthly cost: $3.80
vs All Cloud: $19.00
Savings: 80%

Strategy 2: Model Selection

Choose models by importance:

Tier 1: Critical

Transcription: OpenAI Whisper
Analysis: GPT-4o or Gemini 1.5 Pro
Cost: ~$0.40/hour audio
Quality: ⭐⭐⭐⭐⭐

Tier 2: Important

Transcription: OpenAI Whisper
Analysis: GPT-4o-mini or Gemini Flash
Cost: ~$0.38/hour
Quality: ⭐⭐⭐⭐

Tier 3: Routine

Transcription: Ollama Whisper
Analysis: Gemini Flash
Cost: ~$0.01/hour
Quality: ⭐⭐⭐

Strategy 3: Transcript Pre-Processing

Reduce token usage:

Before Processing:
──────────────────
"Um, so like, you know, we need to, uh,
approve the budget, you know?"
→ 15 tokens

After Cleanup:
──────────────
"We need to approve the budget."
→ 6 tokens

Savings: 60% token reduction

Tools:

• Remove filler words (um, uh, like)
• Consolidate repeated information
• Extract key sections only
• Use summarization first (for very long)

Strategy 4: Batch Processing Windows

Schedule processing during off-peak:

1. Record during day (no processing)
2. Queue all recordings
3. Process overnight (local Ollama)
4. Review results next morning

Benefits:
- Free processing (local)
- No API rate limits
- Batch efficiency
- Results ready by morning

Strategy 5: API Key Optimization

Use separate keys for monitoring:

Development Key:
- Testing and iteration
- Lower spending limit ($10/month)
- Alert if exceeded

Production Key:
- Important transcripts only
- Higher spending limit ($100/month)
- Separate billing tracking

---

Batch Processing Workflows

Workflow 1: Weekly Meeting Batch

Scenario: 10 hours of meetings per week

Process:

Monday-Friday:
1. Record all meetings (auto-transcribe: ON)
2. Transcripts accumulate
3. No analysis yet

Friday evening:
4. Select all week's transcripts
5. Batch analyze with Ollama (free)
6. Review over weekend

Efficiency:

• 10 transcriptions: 20-50 minutes (Ollama)
• Run overnight on Friday
• Zero cost
• Full week insights ready

Workflow 2: Multi-File Upload

Scenario: Import 20 historical transcripts

Process:

1. Prepare all files:
   - Convert to .txt or .docx
   - Clean formatting
   - Name consistently

2. Upload in sequence:
   - Drag-drop multiple files (planned)
   - Or upload one-by-one
   - Let queue process

3. Monitor progress:
   - Check processing status
   - Handle errors individually
   - Export results

Workflow 3: Reprocessing Campaign

Scenario: Improve old transcripts with better models

Process:

1. Identify candidates:
   - Search: date:2024-01 (old month)
   - Filter: processed_with:gpt-3.5-turbo

2. Export originals (backup):
   - Select all
   - Export as ZIP

3. Reprocess batch:
   - Select all
   - Right-click → "Reprocess with..."
   - Choose gpt-4o-mini

4. Compare results:
   - Side-by-side comparison
   - Keep better version

---

Job Queue Management

Feature: F021 - Backend Job Queue System

Monitor and manage background processing.

Queue Status View

┌──────────────────────────────────┐
│ Job Queue Status                 │
├──────────────────────────────────┤
│ Active Jobs: 3                   │
│ Queued: 7                        │
│ Completed: 45                    │
│ Failed: 2                        │
│                                  │
│ Current:                         │
│ • Transcribing meeting_042.webm  │
│   Progress: 67% (2min remaining) │
│                                  │
│ • Analyzing transcript_031.txt   │
│   Progress: 34% (30s remaining)  │
│                                  │
│ [Pause Queue] [Clear Completed]  │
└──────────────────────────────────┘

Queue Operations

Pause processing:

• Finish current jobs
• Hold new jobs in queue
• Resume when ready

Priority adjustment: (Planned)

• Move important jobs to front
• Delay low-priority jobs
• Manual queue reordering

Cancel jobs:

• Remove from queue
• Stop in-progress jobs
• Clean up partial results

Automatic Retry Logic

Failed jobs auto-retry:

Attempt 1: Immediate
Attempt 2: Wait 30 seconds
Attempt 3: Wait 2 minutes
Attempt 4: Wait 5 minutes
Final: Mark as failed

Retry conditions:

• Network timeouts
• Transient API errors
• Rate limit exceeded

Non-retryable:

• Invalid API keys
• Malformed files
• Quota exceeded

---

Keyboard Shortcuts Reference

Global Shortcuts

Shortcut	Action
Ctrl/Cmd + K	Open global search
Ctrl/Cmd + B	Toggle sidebar
Ctrl/Cmd + ,	Open settings (planned)
Ctrl/Cmd + N	New project (planned)
Ctrl/Cmd + U	Upload transcript
/	Focus search

Navigation

Shortcut	Action
↑ ↓	Navigate lists
Enter	Select/open
Esc	Close modal/cancel
Tab	Next field
Shift + Tab	Previous field

Editing

Shortcut	Action
Ctrl/Cmd + Z	Undo
Ctrl/Cmd + Y	Redo
Ctrl/Cmd + S	Save (planned)
Enter	Save edit
Esc	Cancel edit

Recording (Planned)

Shortcut	Action
Ctrl/Cmd + R	Start recording
Ctrl/Cmd + P	Pause recording
Ctrl/Cmd + Shift + R	Stop recording

Creating Custom Shortcuts

Via Settings (Planned):

Settings → Keyboard Shortcuts
    ↓
Customize any action
    ↓
Assign key combination

---

Notification Management

Feature: F019 - Toast Notification System

Manage how you receive feedback.

Notification Types

Success toasts:

✅ Transcript uploaded successfully
✅ Processing complete
✅ Settings saved

Error toasts:

❌ Failed to save changes
❌ API key invalid
❌ Connection timeout

Info toasts:

ℹ️ Processing in background
ℹ️ Update available
ℹ️ Low disk space warning

Notification Settings

┌──────────────────────────────────┐
│ Notification Preferences         │
├──────────────────────────────────┤
│ ☑ Show success messages          │
│ ☑ Show error messages            │
│ ☑ Show info messages             │
│ ☐ Show processing updates        │
│                                  │
│ Duration: [3 seconds        ▼]   │
│ Position: [Top Right        ▼]   │
│ Sound:    [Enabled          ▼]   │
└──────────────────────────────────┘

Notification Actions

Click to navigate:

• Click success toast → Go to transcript
• Click error toast → Open relevant settings
• Click warning → View details

Dismiss:

• Click × to dismiss
• Auto-dismiss after 3-5 seconds
• Swipe to dismiss (mobile)

---

Widget Customization

Feature: F021 - Widget Order Customization

Personalize your visualization layout.

Creating Custom Layouts

Executive Dashboard:

1. Action Matrix (what needs doing)
2. Sentiment Arc (how it went)
3. Decision Flowchart (what was decided)
4. Participation Heatmap (who engaged)
5. Concept Mind Map (what was discussed)

Project Manager View:

1. Action Matrix (task tracking)
2. Participation Heatmap (team engagement)
3. Decision Flowchart (dependencies)
4. Concept Mind Map (scope)
5. Sentiment Arc (team morale)

Analyst View:

1. Sentiment Arc (emotional dynamics)
2. Participation Heatmap (speaking patterns)
3. Concept Mind Map (topic analysis)
4. Decision Embedding (decision clusters)
5. Action Matrix (deliverables)

Per-Project Layouts (Planned)

Different layouts for different projects:

Client Meetings:
- Focus on actions and decisions
- Hide participation (privacy)

Team Retrospectives:
- Emphasize sentiment and participation
- Show concept exploration

Strategy Sessions:
- Feature decision embedding
- Highlight concept relationships

---

Custom Model Integration

Adding Custom Ollama Models

Fine-tuned domain models:

# Create custom model
ollama create meeting-analyzer -f Modelfile

# Modelfile contents:
FROM llama3.1:latest
PARAMETER temperature 0.7
SYSTEM You are an expert meeting analyzer...

Add to Selfoss:

Settings → LLM & Processing
    ↓
Manage Custom Models
    ↓
Provider: Ollama
Model: meeting-analyzer
    ↓
[Add Model]

Custom Prompts (Future)

Specialized extraction:

Template: Technical Meeting
- Extract: Technical decisions, specs, blockers
- Ignore: Social chatter, off-topic

Template: Sales Call
- Extract: Objections, commitments, next steps
- Ignore: Small talk, generic discussion

---

Next Steps

🎉 You’re now a Selfoss power user!

Master Techniques:

1. 🔄 Implement hybrid setup - Balance privacy/cost/speed
2. 💰 Optimize spending - Track and reduce costs
3. ⚡ Automate workflows - Batch processing
4. ⌨️ Learn all shortcuts - Maximum efficiency
5. 🎨 Customize widgets - Perfect layout

Contribute:

• 📝 Share your workflows
• 🐛 Report issues
• 💡 Suggest features
• 🤝 Help others

---

⚡ Work smarter, not harder.