Skip to content

20-minute-workflow.md

Latest commit

 

History

History
202 lines (146 loc) · 5.53 KB

20-minute-workflow.md

File metadata and controls

202 lines (146 loc) · 5.53 KB
Raw

20-Minute Workflow

Fast execution of systematic AI research (after initial setup)

Prerequisites

This workflow assumes you've already completed the discovery process:

  • ✅ Problem clearly defined through alignment phase
  • ✅ Master research prompt created and tested
  • ✅ Research scope narrowed to specific target
  • ✅ Evidence standards established

If you haven't done this setup, see methodology.md for the complete 60-minute discovery process.

Pre-Work (2 minutes)

Define Today's Research Target

Fill out these specific parameters:

Domain: [Your industry/field]
Subject: [Specific thing to research]
Scope: [What you need to know]
Business Need: [Why this matters now]

Example from Purple Case:

Domain: Mattress specifications
Subject: Purple RestorePlus Cool Touch 13" Queen
Scope: Coil count, construction layers, exclusivity verification
Business Need: Customer asking technical questions not in training materials

Set Up Workspace

  • Open 4 browser tabs: ChatGPT, Claude, Gemini, Perplexity
  • Have your master research prompt ready (customized from research-prompt.md)
  • Create document for collecting outputs
  • Start timer

Phase 1: Multi-LLM Research Deployment (8 minutes)

Deploy Your Master Prompt (2 minutes per system)

Run your pre-built research prompt across all systems simultaneously:

All 4 Systems - Parallel Execution:

  1. Paste your customized master prompt into each LLM
  2. Ensure JSON output format is being generated
  3. Copy each result to collection document as received
  4. Label outputs clearly (ChatGPT/Claude/Gemini/Perplexity)

Quality Check: Each output should have:

  • JSON structure
  • Evidence sources with URLs
  • Confidence levels (High/Medium/Low)
  • Technical specifications attempted

Time Check: 8 minutes elapsed

Phase 2: Cross-Analysis (5 minutes)

Find Consensus and Conflicts

Return to your preferred LLM with this analysis prompt:

I have research outputs from 4 different AI systems on [YOUR SUBJECT].
Analyze for consensus and conflicts:

1. CONSENSUS: What facts do 3+ systems agree on?
2. CONFLICTS: What information differs between systems?
3. EVIDENCE QUALITY: Which sources are most authoritative?
4. CONFIDENCE: What can we state with certainty?

[PASTE ALL 4 OUTPUTS]

Focus on HIGH confidence consensus for customer-facing use.

Purple Case Example Output:

  • Consensus: 3" GelFlex Grid (all 4 systems)
  • Conflict: Coil count varies 789-892
  • Best Evidence: Purple.com official specs
  • High Confidence: Height, Grid thickness, exclusivity

Time Check: 13 minutes elapsed

Phase 3: Validation and Clean Output (7 minutes)

Generate Deployment-Ready Data

Quick Validation (3 minutes)

Using the validation framework:

Validate this synthesized research for deployment:
- Customer-safe claims (HIGH confidence only)
- Technical accuracy (consensus findings)
- Evidence chain (source documentation)

[PASTE CONSENSUS FINDINGS]

Mark anything below HIGH confidence as "requires verification"

Final Output Generation (4 minutes)

Create clean, structured output:

Generate final JSON output with:
- Customer-safe facts (HIGH confidence consensus only)
- Requires verification (MEDIUM confidence items)
- Missing information (identified gaps)
- Evidence trail (source URLs for all claims)

Use the structure from example-output.json

Time Check: 20 minutes total

What This Workflow Delivers

In 20 Minutes You Get:

  • Consensus findings from 4 AI systems
  • Evidence-graded specifications
  • Customer-ready information separated from internal-use
  • Full source documentation
  • Structured JSON for future automation

Compared to Manual Research (3 hours):

  • Same accuracy (95%+ match)
  • Better documentation (every claim sourced)
  • Systematic validation (not subjective)
  • Repeatable process (same quality every time)

When to Use Full 60-Minute Process Instead

Use the complete methodology from methodology.md when:

  • First time researching a new domain
  • Creating new master prompts
  • Training team members
  • Establishing evidence standards
  • Discovering optimal research approach

Quality Checkpoints

8-Minute Mark (After Research)

  • Have 4 JSON outputs collected
  • Each has evidence documentation
  • Technical specs attempted by all systems

13-Minute Mark (After Analysis)

  • Consensus findings identified
  • Conflicts documented
  • Confidence levels assigned

20-Minute Mark (Final)

  • Customer-safe facts separated
  • Evidence chain complete
  • Ready for deployment

Troubleshooting Quick Reference

If This Happens Do This
No JSON from some LLMs Add "Return ONLY valid JSON" to prompt
No consensus found Focus on official sources only
Taking >20 minutes Narrow scope to critical specs
Poor evidence quality Explicitly require URLs for all claims

Scaling This Workflow

For Daily Use:

  • Save domain-specific prompts
  • Build library of validated outputs
  • Track which LLMs perform best for your domain
  • Refine evidence standards based on results

For Team Adoption:

  • Share master prompts in team repository
  • Establish consensus thresholds (3/4 systems = valid)
  • Create domain-specific validation criteria
  • Document successful research examples

Remember: This 20-minute workflow is for execution after you've done the groundwork. The first research in any new domain should use the full 60-minute methodology to establish your approach.