OSF Open Source Fund

🏠 OSF - Open Source Fund

AI-driven property tokenization simulation for learning, not investing

> What is OSF?

OSF (Open Source Fund) is an educational simulation that demonstrates how AI could manage a transparent, community-governed property investment network. Built for the Gemini 3 Hackathon (Marathon Agent track), it allows users to explore property tokenization concepts without any financial risk.

Aspect	Details
Type	Educational sandbox
Stack	SvelteKit, FastAPI, SQLite
AI	Gemini 2.0 Flash, Imagen 3
Hackathon	Gemini 3 (Deadline: Feb 9, 2026)
Track	Marathon Agent

⚠️ Note: This is a hackathon project for educational purposes. No real money, no real assets, no financial advice.

> Demo Video

Watch the 3-minute walkthrough of OSF in action:

> Podcast: Simulating a Self-Healing Housing Market

Listen to the deep-dive discussion on OSF's architecture and concepts:

> Architecture

graph TB
    subgraph Frontend["Frontend (SvelteKit)"]
        UI[Simulation Dashboard]
        Clock[Network Clock]
        Chat[Governor Chat]
    end
    
    subgraph Backend["Backend (FastAPI)"]
        API[API Layer]
        Batch[Batch Processor]
        NPC[NPC Manager]
        Events[Event Generator]
    end
    
    subgraph AI["Gemini AI"]
        Flash[Gemini 2.0 Flash]
        Imagen[Imagen 3]
    end
    
    subgraph Data["Data Layer"]
        DB[(SQLite)]
        Pool[Property Pool]
        Avatars[Avatar Pool]
    end
    
    UI --> API
    Clock --> API
    Chat --> Flash
    API --> Batch
    Batch --> Flash
    Batch --> NPC
    NPC --> Events
    Events --> DB
    Imagen --> Pool
    Imagen --> Avatars

> Two-Phase Processing Model

The core innovation is a two-phase architecture that optimizes token usage while maintaining interactivity:

sequenceDiagram
    participant User
    participant Frontend
    participant Backend
    participant Gemini
    
    Note over User,Gemini: Phase 1: Real-Time (Between Ticks)
    User->>Frontend: Ask Governor question
    Frontend->>Gemini: Stream request
    Gemini-->>Frontend: Streaming response
    
    Note over User,Gemini: Phase 2: Batch Tick (Monthly)
    Backend->>Backend: Collect all state
    Backend->>Gemini: Single batch call (~50K tokens)
    Gemini-->>Backend: AI decisions + narrative
    Backend->>Frontend: SSE broadcast
    Frontend->>User: Month completed!

Phase Details

Phase	Model	Context	Purpose
Real-time chat	Gemini Flash	~10K tokens	Governor/Advisor Q&A
Batch processing	Gemini Flash	~50K tokens	Monthly simulation tick
Property images	Imagen 3	N/A	Isometric renders
Character avatars	Imagen 3	N/A	NPC/user avatars

> Marathon Agent Implementation

OSF demonstrates the Marathon Agent pattern through autonomous simulation:

graph LR
    subgraph Thought["Thought Signature Cycle"]
        O[Observation] --> A[Analysis]
        A --> D[Decision]
        D --> Act[Action]
        Act --> R[Reflection]
        R --> O
    end
    
    subgraph Marathon["Marathon Mode"]
        Start[Start] --> Tick[Auto-tick 2s]
        Tick --> NPCs[11 NPCs Decide]
        NPCs --> Log[AI Thinking Tab]
        Log --> Tick
    end

Marathon Mode Features

Feature	Implementation
Duration	120 months (10 years) at 2s/tick ≈ 4 hours
NPCs	11 unique personalities with goal-driven behavior
Self-correction	NPCs adjust risk tolerance based on performance
Visible reasoning	AI Thinking tab shows all decisions live
Session continuity	Tracks elapsed time and total months

> Self-Healing Network

Unlike traditional markets, OSF can detect and recover from stress:

graph TB
    subgraph Detect["Sense & Diagnose"]
        M[Monitor Metrics] --> D[Diagnose Root Cause]
    end
    
    subgraph Respond["Respond & Verify"]
        D --> R[Activate Countermeasures]
        R --> V[Verify Success]
        V --> L[Learn & Improve]
    end
    
    subgraph Strategies["Healing Strategies"]
        S1[Liquidity Pool<br/>Deploy floor bids]
        S2[Buyer-Seller Matching<br/>Connect distressed sellers]
        S3[Partial Exit Program<br/>30% now, 70% over 6mo]
        S4[Rent-to-Own Acceleration<br/>Alert tenants to buy in]
    end
    
    R --> S1
    R --> S2
    R --> S3
    R --> S4

Health Indicators

Metric	Healthy Threshold
Liquidity	>80%
Exit Queue	<5 requests
Trade Failures	<15%
Occupancy	>95%
Rent Collection	>98%

> Explorable Roles

Users can experience property from 6 different perspectives:

Role	Experience
Investor	Buy/sell tokens, track portfolio, vote on proposals
Renter	Select property, pay rent, request maintenance
Tenant	Rent-to-own pathway, equity accumulation
Homeowner	Equity access, rental income, property listing
Service Provider	Complete tasks as plumber, electrician, manager
Foundation	Stake tokens, enhanced governance, yields

Connected Roles

When a homeowner accesses equity, their holdings appear in the investor view:

Homeowner accesses $50K equity
        ↓
Receives OSF tokens (marked "Your Home Equity")
        ↓
Tokens appear in Investor portfolio
        ↓
Tokens earn dividends like regular investments

> Cooperative Gamification

OSF intentionally avoids competitive mechanics:

Traditional Games	OSF Approach
Beat other players	Contribute to network health
Maximize personal wealth	Support collective outcomes
Individual rankings	Network health grade (A-F)
Competitive leaderboards	Exploration achievements

Collective Outcomes

Outcome	What It Measures
Families Housed	Stable housing provided
Dividends Distributed	Shared with token holders
Equity Accessed	Homeowners helped
Crises Survived	Downturns weathered
Evictions	0 (OSF prevents this)
Foreclosures	0 (cooperative model)

Exploration Achievements

First Investment - Make your first token purchase
Weathered Storm - Hold during a market downturn
Witnessed Boom - Experience a boom market
Saw Self-Healing - Watch the network recover
Full Cycle - Complete a full boom-bust cycle
Network Anchor - Maintain holdings during crisis

> WA Market Dynamics

The simulation is calibrated to Western Australian property data:

Metric	Value	Source
Median house price (Perth)	~$750,000	REIWA
Rental vacancy rate	<1%	REIWA
Gross rental yield (houses)	~4.5%	CoreLogic
Gross rental yield (units)	~6.0%	CoreLogic
Population growth	~2.3%	WA Treasury

Boom-Bust Cycle

Condition	Monthly Appreciation	Triggers
Boom	+0.6% to +1.2%	Iron ore >$150/t, pop growth >2%
Stable	+0.2% to +0.4%	Iron ore >$100/t
Stagnant	-0.1% to +0.1%	Extended peak, buyer hesitation
Declining	-0.5% to -0.2%	Iron ore <$100/t
Bust	-1.0% to -0.5%	Iron ore <$80/t, population outflow

> Tech Stack

Layer	Technology
Frontend	SvelteKit 2.0, Svelte 5 Runes, TypeScript
UI Components	shadcn-svelte, Lucide icons, Chart.js
Interactive Diagrams	@xyflow/svelte (Money Flow, Self-Healing)
Backend	FastAPI, Python 3.11+, SQLAlchemy 2.0
Database	SQLite (demo) / PostgreSQL (production)
AI	Gemini 2.0 Flash (streaming), Imagen 3
Real-time	Server-Sent Events (SSE)
Market Data	RBA, ABS, APRA, REIWA, PropTrack

> API Overview

Clock Endpoints

# Get clock status
GET /network/clock/status

# Start marathon mode
POST /network/clock/start

# Force tick (debug)
POST /network/clock/force-tick

# SSE event stream
GET /network/clock/stream

AI Endpoints

# Chat with Governor (streaming)
GET /network/governor/chat/stream?message=How%20is%20the%20network?

# Portfolio advice
POST /network/advisor/portfolio

> Key Innovations

1. Batch Context Maximization

Single Gemini call processes entire monthly state instead of hundreds of individual requests.

2. Emergent NPC Dynamics

11 NPCs with unique personalities create realistic market behavior without scripting.

3. Self-Healing Mechanisms

Network can coordinate participants to solve problems traditional markets cannot.

4. Cooperative Design

Rewards network health and learning, not wealth accumulation or competition.

5. Connected Role State

Homeowner actions flow through to investor portfolios, demonstrating real token economics.

> Running Locally

# Backend
cd projects/osf-demo/backend
export GOOGLE_API_KEY=your_key
USE_SQLITE=true uvicorn src.main:app --port 8000

# Frontend
cd projects/osf-demo/frontend
npm run dev

> Project Status

This is a hackathon project submitted for the Gemini 3 Hackathon (Marathon Agent track). It demonstrates AI capabilities in autonomous, long-running simulations with cooperative mechanics.

What it is:

Educational simulation
Hackathon demo
Proof of concept

What it is NOT:

Financial product
Investment advice
Production system

> Links

Track: Marathon Agent
Prize Pool: $100,000
Deadline: February 9, 2026

OSF - AI-Powered Property Network Simulation