Skip to main content

Overview

Dream Foundry transforms founder ideas into production code through five sequential phases. Each phase has specific responsibilities and outputs that feed into the next.
┌─────────────┐  ┌─────────────┐  ┌─────────────┐  ┌─────────────┐  ┌─────────────┐
│    THE      │  │    THE      │  │    THE      │  │    THE      │  │    THE      │
│ DREAMCATCHER│─▶│DREAM FACTORY│─▶│ DREAM ARENA │─▶│DREAM PODIUM │─▶│  DREAM      │
│             │  │             │  │             │  │             │  │ AWAKENING   │
│ Innovative  │  │ AI Agent    │  │ Build, Run  │  │ Polished by │  │ Feature     │
│ Ideas,      │  │ Candidates, │  │ & Measure,  │  │ CodeRabbit, │  │ Presentation│
│ Objectives  │  │ Targets Set │  │ Top Three   │  │ Internal    │  │ Release to  │
│ & Constraints│ │ with Bias   │  │ Chosen      │  │ Testing     │  │ the World   │
└─────────────┘  └─────────────┘  └─────────────┘  └─────────────┘  └─────────────┘

Phase 1: The Dreamcatcher

Goal: Transform raw founder vision into structured requirements

What Happens

1

Idea Capture

Founder inputs their raw idea, e.g., “I want a bot that posts AI events to our Discord every week”
2

Context Extraction

System captures:
  • Context: Where the idea came from (CES, customer request, competitive pressure)
  • Aspirations: What success looks like
  • Constraints: Budget, timeline, tech stack limitations
3

Brief Generation

Creates a structured IdeaBrief that serves as the foundation for all subsequent phases

Output

An IdeaBrief object containing:
  • Structured description
  • Success criteria
  • Hard constraints (time, budget, tech)
  • Business context
The Dreamcatcher phase is about clarity, not implementation. It forces founders to articulate what they actually want before any code is written.

Phase 2: The Dream Factory

Goal: Generate multiple competing implementation approaches

What Happens

1

Candidate Generation

AI generates N different implementation approaches, each with distinct trade-offs
2

Objective Setting

For each candidate, define measurable objectives with bias weights:
  • H (High): 3x multiplier - Critical requirements
  • M (Medium): 2x multiplier - Important but not critical
  • L (Low): 1x multiplier - Nice to have
3

Scaffold Creation

Generate starter code for each candidate approach

Demo Agents

The hackathon demo includes 5 pre-built agents showcasing different strategies:

Agent Alpha

The Speed DemonLightning fast scraper that prioritizes speed over completeness. Skips weekend events and only checks top sources.Trade-off: Fast execution, but may miss important events

Agent Beta

The PerfectionistThorough scraper that verifies every URL with full HTTP requests. Includes comprehensive event coverage.Trade-off: Slower execution, but higher quality output

Agent Gamma

The InsiderUses curated data from Cerebral Valley and verified lu.ma sources. Reliable connections, fast execution.Trade-off: Limited to known sources, but highly reliable

Agent Delta

The CrasherIntentionally crashes with divide-by-zero error to demonstrate Sentry error capture.Purpose: Shows how the system handles failures gracefully

Agent Epsilon

The HallucinatorProduces intentionally bad data (wrong dates, wrong locations, fake events) to demonstrate quality scoring.Purpose: Validates that the scoring system catches low-quality output

Output

A list of Candidate objects, each containing:
  • Implementation approach description
  • Tech stack selection
  • Complexity estimate
  • Starter code scaffold
  • Measurable objectives with weights

Phase 3: The Dream Arena

Goal: Build, run, and measure all candidates to find the top 3

What Happens

1

Daytona Sandbox Creation

Each candidate gets an isolated Daytona workspace
# From forge.py:242
daytona_results = run_all_candidates_in_daytona(
    CANDIDATES, objective, artifacts_dir, log_callback
)
2

Sentry Monitoring

Sentry captures errors and performance metrics during execution
# From forge.py:176
sentry_sdk.capture_message(
    f"Candidate {candidate_id} failed: {error_message[:200]}",
    level="error"
)
3

Execution

Each candidate runs with:
  • 60-second timeout
  • Error capture
  • Performance tracking
  • Artifact generation
4

Scoring

Candidates scored on three criteria:
  • Success (20%): Did it produce output without errors?
  • Quality (60%): Does output meet requirements?
  • Speed (20%): How fast did it execute?
5

Top 3 Selection

Highest-scoring candidates advance to the Podium

Key Demo Moment

Sentry captures Agent Delta’s crash in real-time, demonstrating error monitoring:
# From agent_delta.py:23
events_found = 0
average = 100 / events_found  # BOOM! DivisionError

Output

  • scores.json: Complete scoring breakdown for all candidates
  • winner.txt: ID of the winning candidate
  • Artifacts for each candidate (Discord posts, logs)
  • Sentry error reports

Phase 4: The Dream Podium

Goal: Polish the top 3 candidates and select the final winner

What Happens

1

CodeRabbit Review

Top 3 candidates submitted to CodeRabbit for AI-powered code review
2

Code Improvements

CodeRabbit suggests and applies:
  • Bug fixes
  • Performance optimizations
  • Code quality improvements
  • Best practices enforcement
3

Re-testing

All polished candidates re-run through the same test suite
4

Before/After Comparison

System shows improvement deltas:
  • Score changes
  • Error reduction
  • Performance gains
5

Winner Selection

Highest-scoring polished candidate wins
6

PR Publication

Winner’s code published as a pull request

Key Demo Moment

Agent Alpha jumps from 3rd to 2nd place after CodeRabbit adds weekend event coverage, demonstrating measurable improvement.

Output

  • Polished code for top 3 candidates
  • Before/after diffs
  • Updated scores
  • Winner’s pull request

Phase 5: The Dream Awakening

Goal: Present the feature to the world

What Happens

1

Documentation Generation

Generate three types of docs:
  • Engineering: Technical implementation details
  • Marketing: Feature benefits and positioning
  • Executive: Business impact and ROI
2

Voice Narration

ElevenLabs generates audio presentations for each audience
# From config.yaml
showcase:
  voice_model: "eleven_multilingual_v2"
  output_formats: ["mp3"]
3

Release Package

Bundle everything:
  • Production code
  • Documentation
  • Audio presentations
  • Demo videos
4

Launch

Feature goes live:
  • Discord announcement
  • Team notification
  • Stakeholder updates

Key Demo Moment

ElevenLabs narration summarizes the entire journey and announces Agent Gamma as the winner.

Output

  • Complete feature documentation
  • MP3 audio presentations
  • Marketing materials
  • Launch announcement
  • Production deployment

How each sponsor tool integrates across phases:
PhaseDaytonaSentryCodeRabbitElevenLabs
Dreamcatcher
Dream Factory
Dream Arena⭐ Sandboxes⭐ Monitoring
Dream Podium⭐ Polish + PR
Dream Awakening⭐ Presentation

Complete Journey Example

Founder’s Request: “I want a bot that posts AI events to our Discord every week.”
Captures requirement:
  • Goal: Weekly AI events in SF for Discord
  • Success: 10+ events, includes Daytona hackathon, valid URLs
  • Constraints: Must run in under 30 seconds
Generates 5 agents:
  • Alpha: Fast scraper (weekday events only)
  • Beta: Thorough scraper (all events, slower)
  • Gamma: Curated sources (verified lu.ma data)
  • Delta: Intentional crasher (Sentry demo)
  • Epsilon: Bad data generator (quality demo)
All 5 compete:
  • Delta crashes immediately (Sentry captures)
  • Epsilon produces bad data (quality score: 12/100)
  • Alpha is fast but misses hackathon (quality: 45/100)
  • Beta finds all events but slow (quality: 88/100)
  • Gamma wins: fast + comprehensive (quality: 92/100)
CodeRabbit polishes top 3:
  • Alpha gets weekend event fallback → jumps to 72/100
  • Beta gets performance optimizations → stays at 91/100
  • Gamma gets minor cleanups → stays at 94/100
Winner: Agent Gamma (94/100 total score)
ElevenLabs presents:
  • Engineering doc: Implementation details
  • Marketing pitch: “Never miss an AI event in SF”
  • Executive summary: “Automated community engagement”
Feature ships to production ✅

Next Steps

Scoring System

Learn how the Success/Quality/Speed scoring works

Agent Strategies

Deep dive into each agent’s implementation approach

Build docs developers (and LLMs) love