Skip to content

Scenario 4: Conference Workshop

Persona: Dr. Alex Rivera - Workshop Instructor

Background: - Assistant Professor, Machine Learning researcher - Accepted to teach 3-hour workshop at NeurIPS 2025 - Workshop: "Hands-on Deep Learning with PyTorch" - Expected attendance: 40-60 participants (international) - Budget: $200 from conference organizers (one-time allocation) - Critical constraint: Must work perfectly on first try - no second chances

Pain Points: - Participants arrive with varying laptop configurations (Windows/Mac/Linux) - Limited time for troubleshooting (workshop starts in 90 minutes) - Need identical environments for all participants to follow along - Budget must cover entire workshop duration + buffer - International participants in multiple timezones for pre-workshop prep - Must auto-terminate - can't rely on participants to clean up afterwards

Workshop Structure: - Week before: Send invitation links to registered participants - Day before: Early access for testing (24-hour window) - Workshop day: 3-hour hands-on session - Auto-cleanup: Terminate all workspaces 3 hours after workshop ends

Current State (v0.5.5): What Works Today

✅ Pre-Workshop Setup (1 Week Before)

# Alex sets up workshop environment
cws profile create neurips-workshop --aws-profile alex-research --region us-west-2

# Create template-restricted project for workshop
cws project create neurips-dl-workshop \
  --budget 200 \
  --description "NeurIPS 2025: Deep Learning Workshop" \
  --alert-threshold 80

# Generate batch invitations for 60 participants
cat > workshop_participants.csv << EOF
Name,Type,ValidDays,CanInvite,Transferable,DeviceBound,MaxDevices
Participant_01,read_only,7,no,no,yes,2
Participant_02,read_only,7,no,no,yes,2
Participant_03,read_only,7,no,no,yes,2
[... 57 more participants ...]
EOF

# Create invitations with basic policy restrictions
cws profiles invitations batch-create \
  --csv-file workshop_participants.csv \
  --output-file invitation_codes.csv \
  --include-encoded

Current capabilities: - ✅ Batch invitation generation (60 participants in seconds) - ✅ Time-boxed access (7-day expiration) - ✅ Access extension: Can extend for additional day(s) so participants can continue working - ✅ Device binding (prevents casual sharing) - ✅ Budget allocation ($200 total) - ✅ Basic policy restrictions (template whitelist)

💡 Workshop Extension Example: After 3-hour workshop ends, Alex can extend access for 24 hours: 

cws profiles invitations extend neurips-workshop --add-days 1
# All 60 participants get automatic 24-hour extension
# Great for: Homework completion, extended tutorials, follow-up work

✅ Day Before Workshop (Early Access Testing)

# Participants receive email with invitation link
# They accept invitation and test their environment

# Participant workflow:
cws profiles invitations accept <INVITATION-CODE> neurips-workshop
cws launch pytorch-ml workshop-test --size S

# Alex monitors early access
cws project workspaces neurips-dl-workshop
# Output:
# ✅ 12 participants tested successfully
# ⚠️  3 participants having issues (Alex contacts them)
# 💰 Current spend: $4.20 (within budget)

⚠️ Current Pain Points: What Doesn't Work

❌ Problem 1: No Automatic Workspace Termination

Scenario: Workshop ends at 3:00 PM, workspaces should terminate at 6:00 PM

What should happen (MISSING): 

# Alex launches workspaces with auto-terminate timer
cws launch pytorch-ml workshop-instance --hours 6

# CloudWorkstation output:
# ✅ Workspace launching: workshop-instance
# ⏰ Auto-terminate scheduled: 6 hours from now (6:00 PM)
# 📊 Cost for 6 hours: $3.20
# 🔔 Warning will be sent 30 minutes before termination

Current workaround: Alex must manually stop 60 workspaces or rely on participants Risk: If forgotten, $200 budget exhausted in 3 days

❌ Problem 2: No Template Whitelisting at Invitation Level

Scenario: Participants should ONLY be able to launch PyTorch ML template

What should happen (MISSING): 

# Create invitations with template restrictions
cws profiles invitations batch-create \
  --csv-file participants.csv \
  --template-whitelist "PyTorch Machine Learning" \
  --max-instance-type "t3.medium" \
  --output-file invitations.csv

# When participant tries wrong template:
participant$ cws launch gpu-ml-workstation expensive-instance
# ❌ Error: Template 'gpu-ml-workstation' not allowed by your invitation policy
#    Allowed templates: ["PyTorch Machine Learning"]
#
#    This is a workshop environment with restricted templates.
#    Please use: cws launch "PyTorch Machine Learning" my-instance

Current workaround: Trust participants + budget alerts Risk: Single participant launches GPU workspace → $600/day → budget blown in 8 hours

❌ Problem 3: No Bulk Launch for Pre-Provisioning

Scenario: Workshop starts at 9:00 AM, Alex wants all environments ready at 8:45 AM

What should happen (MISSING): 

# Night before workshop: Pre-provision all instances
cws project bulk-launch neurips-dl-workshop \
  --template "PyTorch Machine Learning" \
  --count 60 \
  --name-pattern "workshop-{01-60}" \
  --start-time "2025-12-08T08:45:00" \
  --terminate-hours 6

# Output:
# 🚀 Scheduling 60 workspace launches for Dec 8, 8:45 AM
# 📊 Estimated cost: $192.00 (within $200 budget ✅)
# ⏰ All workspaces will auto-terminate at 2:45 PM (3-hour workshop)
#
# 💡 Effective Cost Analysis:
#    24/7 assumption: $2.40/hour × 60 workspaces × 24 hours = $3,456
#    Actual workshop cost: $2.40/hour × 60 workspaces × 3 hours = $432
#    Your cost with auto-terminate: $192 (early terminations banked immediately)
#    Savings: $240 banked in real-time as participants finish early!
#
# Workspace name assignments:
# - Participant_01 → workshop-01
# - Participant_02 → workshop-02
# ...

# 8:45 AM on workshop day - all workspaces auto-launch
# 9:00 AM - participants arrive, workspaces are ready

💡 GUI Note: Workshop scheduling available in GUI Projects tab with calendar view - coming soon in v0.6.0

Current workaround: Participants launch on-demand (slow, error-prone) Impact: First 30 minutes wasted on environment setup

❌ Problem 4: No Real-Time Workshop Dashboard

Scenario: During workshop, Alex needs to see participant progress at a glance

What should happen (MISSING): 

cws workshop dashboard neurips-dl-workshop

# Terminal dashboard (live updates):
# ┌─────────────────────────────────────────────────────────┐
# │ NeurIPS DL Workshop - Live Dashboard                   │
# │                                                         │
# │ Participants:     58 / 60 active                       │
# │ Instances:        58 running, 2 stopped                │
# │ Avg Uptime:       1h 23m (82 compute hours total)     │
# │                                                         │
# │ Budget:          $38.40 / $200.00 (19%) ✅            │
# │ Available:       $161.60 (real-time as terminations happen) │
# │ Effective cost:  $0.47/hour (vs $2.40/hour 24/7)     │
# │                                                         │
# │ 💡 Real-time banking: 2 early finishers already banked $4.80! │
# │ Time Remaining:   1h 37m until auto-terminate          │
# │                                                         │
# │ Participants Needing Help:                             │
# │   ⚠️  workshop-27: Workspace stopped (needs restart)    │
# │   ⚠️  workshop-43: High error rate (check logs)        │
# │                                                         │
# │ Cost by Status:                                         │
# │   Running:  $38.40/hr (58 instances)                   │
# │   Stopped:  $0.00/hr (2 instances)                     │
# │                                                         │
# │ Refresh: Every 30s | Press 'q' to quit                 │
# └─────────────────────────────────────────────────────────┘

💡 GUI Note: Live workshop dashboard available in GUI with real-time participant status - coming soon in v0.6.0

Current workaround: Manual cws list + cws project instances polling Impact: Can't proactively help struggling participants

❌ Problem 5: No Post-Workshop Data Preservation

Scenario: Participants want to keep their workshop code after workspaces terminate

What should happen (MISSING): 

# 30 minutes before auto-terminate, participants receive email:
#
# Subject: ⏰ Workshop Workspace Terminating in 30 Minutes
#
# Your workshop workspace will terminate at 6:00 PM (in 30 minutes).
#
# To preserve your work:
#
# 1. Download your notebook:
#    cws download workshop-instance ~/workshop-code.zip
#
# 2. Or snapshot your instance:
#    cws snapshot create workshop-instance my-workshop-work
#    (This will create a personal AMI - $2.50/month storage)
#
# After termination, you can recreate your environment:
#    cws launch-from-snapshot my-workshop-work restored-env

# Bulk download (instructor):
cws workshop export-all neurips-dl-workshop \
  --output-dir ./participant-work/ \
  --format zip

# Creates:
# ./participant-work/
#   ├── workshop-01.zip (Participant_01's notebooks)
#   ├── workshop-02.zip (Participant_02's notebooks)
#   ...

Current workaround: Participants manually SCP files (most don't) Impact: Lost learning artifacts, can't reproduce workshop results

🎯 Ideal Future State: Complete Workshop Walkthrough

Week Before Workshop: Setup with Auto-Terminate

# Create workshop project with aggressive cost controls
cws project create neurips-dl-workshop \
  --budget 200 \
  --hard-cap \
  --alert-threshold 50,75,90 \
  --description "NeurIPS 2025 Workshop: Deep Learning with PyTorch"

# Create policy-restricted invitations
cws profiles invitations batch-create-workshop \
  --csv-file participants.csv \
  --template-whitelist "PyTorch Machine Learning" \
  --max-instance-type "t3.medium" \
  --max-hourly-cost 0.10 \
  --valid-days 7 \
  --auto-terminate-hours 6 \
  --output-file invitation_codes.csv

# CloudWorkstation output:
# 📧 Generated 60 workshop invitations
#    - Valid for 7 days (expires Dec 9, 2025)
#    - Template restricted: "PyTorch Machine Learning" only
#    - Max instance: t3.medium ($0.0416/hr)
#    - Auto-terminate: 6 hours after launch
#    - Device limit: 2 devices per participant
#
# 📊 Projected costs:
#    - Per participant: $3.20 (6 hours × $0.0416/hr × 1.3 buffer)
#    - Total if all 60 launch: $192.00 ✅ (within $200 budget)
#
# ✅ Invitations saved to: invitation_codes.csv
#
# Next steps:
#   1. Email invitation codes to participants
#   2. Enable early access (optional): cws workshop early-access enable
#   3. Monitor signups: cws workshop participants neurips-dl-workshop

# Email invitation codes to participants
cws workshop email-invitations \
  --csv-file invitation_codes.csv \
  --template workshop_welcome.html \
  --subject "NeurIPS 2025: Deep Learning Workshop Access"

Day Before Workshop: Early Access Testing

# Enable early access window (24 hours before workshop)
cws workshop early-access neurips-dl-workshop \
  --enable \
  --duration 24h \
  --test-mode

# Participants who test early (optional for them):
participant$ cws profiles invitations accept <CODE> neurips-workshop
participant$ cws launch "PyTorch Machine Learning" test-env --hours 2
# (Automatically terminates after 2 hours)

# Alex monitors early access
cws workshop participants neurips-dl-workshop

# Output:
# 📊 Early Access Status (24 hours before workshop)
#
# Accepted Invitations: 58 / 60 (97%)
# Tested Environment:   15 / 58 (26%)
#
# ✅ Ready: 15 participants (tested successfully)
# 🟡 Accepted but not tested: 43 participants
# ❌ Not yet accepted: 2 participants
#    - Participant_23: Invitation sent, not accepted
#    - Participant_47: Invitation sent, not accepted
#
# 💰 Early access cost: $3.20 (15 participants × $0.21/test)
# 📧 Reminder emails:
#    - Send reminder to 43 accepted-not-tested? [Y/n]: y
#    - Send urgent reminder to 2 not-accepted? [Y/n]: y

Workshop Day: Smooth Execution

8:45 AM - Pre-provisioning (optional): 

# Option A: Let participants launch on-demand (default)
# - Slower but gives participants control
# - Launch time: ~2 minutes per instance

# Option B: Pre-provision all workspaces (advanced)
cws workshop bulk-provision neurips-dl-workshop \
  --template "PyTorch Machine Learning" \
  --size S \
  --auto-terminate-hours 6

# Output:
# 🚀 Provisioning 58 workspaces for accepted participants...
# ⏰ Auto-terminate: 6 hours from now (2:45 PM)
#
# Progress: [████████████████████] 58/58 complete (3m 12s)
#
# ✅ All workspaces ready!
# 💰 Current cost: $0.22 (15 minutes of provisioning)
# 📧 Email sent to all participants with connection info

9:00 AM - Workshop begins: 

# Alex opens live dashboard in separate terminal
cws workshop dashboard neurips-dl-workshop --live

# Participants launch (if not pre-provisioned):
participant$ cws launch "PyTorch Machine Learning" workshop-instance
# ✅ Workspace ready in 90 seconds!
# 📓 Jupyter Lab: http://54.123.45.67:8888 (token: abc123)
# ⏰ Workspace will auto-terminate at 3:00 PM (6 hours)
# 💡 To save your work: cws download workshop-instance ~/my-work.zip

10:30 AM - Participant needs help: 

# Dashboard shows participant_27 with high error rate
# Alex remotely debugs (with participant permission):
alex$ cws workshop debug neurips-dl-workshop workshop-27

# Options:
# 1. View Jupyter logs
# 2. View terminal output
# 3. SSH access (requires participant approval)
# 4. Reset notebook kernel
# 5. Restart instance

# Alex selects option 1, identifies issue, helps participant

2:30 PM - 30 minutes before auto-terminate: 

# All participants automatically receive email + terminal notification:
#
# ⏰ Your workshop workspace will terminate in 30 minutes!
#
# Save your work now:
#   cws download workshop-instance ~/neurips-workshop.zip
#
# Or create a snapshot to continue later:
#   cws snapshot create workshop-instance my-dl-work
#   (Costs $2.50/month, can recreate anytime)

# Participants who want to continue (personal budget):
participant$ cws snapshot create workshop-instance my-workshop
# ✅ Snapshot created: my-workshop
# 💰 Storage cost: $2.50/month (personal account)
#
# To recreate:
#   cws launch-from-snapshot my-workshop continued-work

3:00 PM - Workshop ends, auto-terminate begins: 

# CloudWorkstation automatically:
# 1. Sends final warning (5 minutes before)
# 2. Terminates all workspaces at 3:00 PM sharp
# 3. Generates cost report
# 4. Archives workshop data (optional)

# Alex receives final report:
cws workshop report neurips-dl-workshop --export-pdf

# Output:
# 📊 NeurIPS 2025 Deep Learning Workshop - Final Report
#
# Participants:     58 / 60 registered (97%)
# Active instances: 58 workspaces for 3 hours
# Total uptime:     174 instance-hours
#
# Budget:
#   Allocated: $200.00
#   Spent:     $187.45 ✅ (within budget)
#   Saved:     $12.55 (available for next workshop - rollover enabled)
#
#   💡 Effective Cost Analysis:
#      24/7 assumption: $2.40/hr × 58 workspaces × 24 hours = $3,345.60
#      Actual workshop: $2.40/hr × 58 workspaces × 3 hours = $418.00
#      Your actual cost: $187.45 (early terminations banked immediately!)
#      Real-time banking: Every participant who finished early freed budget
#
#   Breakdown:
#     - Workspace compute: $172.90 (58 × 3hrs × $0.99/hr)
#     - Early access:     $3.20 (15 tests)
#     - Pre-provisioning: $0.22 (15min buffer)
#     - Storage:          $11.13 (EBS, snapshots)
#
#   💡 Cloud vs Traditional:
#      Conference room PCs: $60,000 upfront + maintenance
#      CloudWorkstation: $187.45 for 3 hours of actual use
#      You only paid for compute time, not ownership!
#
# Participant Engagement:
#   - High engagement: 42 participants (72%)
#   - Medium engagement: 12 participants (21%)
#   - Low engagement: 4 participants (7%)
#
# Data Preservation:
#   - Snapshots created: 12 participants
#   - Downloads completed: 31 participants
#   - No action: 15 participants (work lost)
#
# ✅ All workspaces terminated successfully
# 💰 No ongoing costs
# 📧 Post-workshop survey sent to all participants

💡 GUI Note: Workshop reports with charts and PDF export available in GUI Reports tab - coming soon in v0.6.0

📋 Feature Gap Analysis

Critical Missing Features (Blockers)

Feature Priority User Impact Current Workaround Effort
Auto-Terminate Timer 🔴 Critical Prevents budget overruns Manual cleanup Medium
Template Whitelisting in Invitations 🔴 Critical Prevents expensive launches Trust + alerts Low
Policy-Restricted Invitations 🔴 Critical Enforces workshop constraints Manual monitoring Medium
Bulk Workspace Provisioning 🟡 High Saves 30min setup time On-demand launch Medium
Workshop Dashboard 🟡 High Real-time participant monitoring Manual polling High

Nice-to-Have Features (Enhancers)

Feature Priority User Impact Benefit
Participant Progress Tracking 🟢 Medium Identify struggling participants Proactive help
Bulk Download/Export 🟢 Medium Preserve participant work Learning continuity
Pre-Workshop Testing 🟢 Medium Catch issues early Smoother workshop
Snapshot Quick-Save 🟢 Low Easy work preservation Student satisfaction
Workshop Templates 🟢 Low Reusable configurations Faster setup

🎯 Priority Recommendations

Phase 1: Workshop Safety Net (v0.7.0)

Target: Workshops can run without budget disasters

  1. Auto-Terminate Timer (1 week)
  2. cws launch template name --hours 6
  3. Countdown warnings at 30min, 5min

  4. Graceful termination with EBS preservation

  5. Invitation Policy Restrictions (1 week)

  6. Template whitelist in invitation tokens
  7. Workspace type restrictions
  8. Hourly cost limits

  9. Policy validation before launch

  10. Workshop Project Type (3 days)

  11. cws project create workshop --type workshop
  12. Built-in auto-terminate defaults
  13. Aggressive budget alerts
  14. One-time budget (no rollover)

Phase 2: Workshop Management Tools (v0.7.1)

Target: Instructors can manage workshops effectively

  1. Workshop Dashboard (1 week)
  2. Live participant status
  3. Real-time budget tracking
  4. Problem detection (stopped instances, errors)

  5. Terminal-based (TUI) interface

  6. Bulk Provisioning (1 week)

  7. Pre-launch workspaces for all participants
  8. Scheduled start time
  9. Coordinated auto-terminate
  10. Assignment to invitation tokens

Phase 3: Workshop Polish (v0.8.0+)

Target: Professional workshop experience

  1. Work Preservation (3 days)
  2. One-click download before terminate
  3. Quick snapshot creation

  4. Bulk export for instructors

  5. Workshop Templates (3 days)

  6. Reusable workshop configurations
  7. Import participant list
  8. One-command workshop setup

Success Metrics

User Satisfaction (Alex's Perspective)

  • ✅ Reliability: "Zero budget disasters - workshop stayed under $200"
  • ✅ Ease of Setup: "60 participants onboarded in 15 minutes"
  • ✅ Peace of Mind: "Auto-terminate means I can focus on teaching, not cleanup"
  • ✅ Participant Success: "97% completion rate - everyone could follow along"

Technical Metrics

  • Auto-terminate prevents 100% of budget overruns
  • Workshop setup time: < 30 minutes (vs 2+ hours manual)
  • Participant environment ready: < 2 minutes (vs 30+ minutes with local install)
  • Zero workspaces left running post-workshop

Business Impact

  • Conference Adoption: "CloudWorkstation workshops" become a standard
  • Reduced Support: Instructors handle workshops independently
  • Positive Reviews: "Best hands-on workshop I've attended!" - Participants
  • Academic Reputation: CloudWorkstation seen as workshop-ready platform

Key Differences from University Class Scenario

Aspect Workshop (3 hours) Class (15 weeks)
Duration Single 3-hour session 15-week semester
Preparation 1 week (must be perfect) 2-4 weeks (iterate)
Budget One-time $200 Semester $1,200 with rollover
Access 6-hour window + cleanup Ongoing with extensions
Cleanup Immediate auto-terminate Gradual semester-end
Support On-site only (3 hours) Office hours + TAs
Participants 40-60 attendees 50 students
TA Structure None or single assistant Head TA + multiple TAs
Failure Cost Workshop disaster Grade assignment issues

Reusable Infrastructure from Class Scenario

✅ Already applicable: - Batch invitation system - Device binding security - Budget allocation and tracking - Template restrictions via policy

🔧 Needs adaptation: - Time limits: 6 hours vs 15 weeks - Budget model: One-time vs recurring - Auto-cleanup: Immediate vs gradual - Support structure: Self-service vs TA hierarchy

Next Steps

  1. Validate with Real Workshop Instructors: Interview 2-3 conference workshop presenters
  2. Prototype Auto-Terminate: Implement basic time-limited launches
  3. Design Workshop Dashboard: Mock up live monitoring interface
  4. Implementation Plan: Break down into 2-week sprints

Estimated Timeline: Workshop Safety Net (Phase 1) → 3 weeks of development

Comparison: Workshop vs Class

Similarities: - Batch user onboarding - Template standardization - Budget constraints - Time-boxed access

Critical Differences: 

Workshop = "High-stakes, single-shot performance"
Class = "Ongoing management with iteration opportunities"

Workshop auto-terminate = "6 hours hard deadline"
Class semester end = "Graceful 2-week wind-down"

Workshop budget = "$200 total, must not exceed"
Class budget = "$1,200 with weekly monitoring and adjustments"

When to use each: - Workshop project: Single-day events, tutorials, short courses - Class project: Semester-long courses, research bootcamps, training programs