unparty.app

unparty demo

UNPARTY is an intelligent context wrapper. Where each component is self-contained, stateless, and intentionally designed to add context to inference calls from the user to the LLM.

The User Layer - Persistence

The user layer provides its own persistence. The user creates an entry. An entry can be a journal - natural language entry.

From there the entry is stored locally and processed through NLP services for inference clustering.

Each component of the app is self-contained. The design is inspired by a MVVM architecture, where the model is represented by the user bucket, view model represented by unparty, and the view represented by the LLM.

  • The User as the Model: The user provides the raw input—their thoughts, ideas, and curiosity—which drives the interaction.

When the context threshold is hit an inference call is made to UNPARTY

  • UNPARTY as the ViewModel: Acting as the intermediary, UNPARTY processes and organizes the user’s input, transforming it into an inference call using machine learning.

The inference call is returned to the user in the form of a prompt to gather further context

  • The LLM as the View: The LLM delivers meaningful responses and visualizations based on the inference call provided by UNPARTY.

Each idea acts as a contextual benchmark. Ideas act as the first step of a component. There are static benchmarks-milestones, that represent a switch. Like a context indicator. The model does not trigger until that context marker is filled. Which triggers the next step. Think of that article as a benchmark for that particular philosophy or foundational step in development. The initial quote was something I encountered in real life that I wrote as a note on a piece of paper. From there I wanted to learn more so I started a prompting session with a LLM. During the session there was an indicator that I did not fully understand the theory or showed that it triggered another idea (important to note you will never be able to understand full intent). The prompting session was saved and went through some python scripts to save metadata and a few others. When I was looking through my collection of notes in my file cabinet, I was reminded of Whitehead. I found the files on my computer, read through the chat, found a missing piece of the current version of unparty which was how can sentiment analysis be used to mark context and not user intent. for the purpose of building the unparty platform sentiment is seen as a natural consequence. consequence being a change - for the model it does not need to remember how it got there, just whats next.

data showed me that people have a trust deficit right now. makes collaboration a bit difficult on top of having busy lives and just general burnout from the constant content pushed in our faces. the quality of ideas coming through social media channels does not match the statistics of entrepreneurship and small business filings. thought a decentralized approach would be best as the app should grow with the user. The best part for the user, is that they can call it whatever they want. unparty was just the foundation. they can build their own and remove it when they are ready.

individual ideas. each idea is a self-contained component of the original idea

graph TD
    subgraph User Layer
        UI[User Interface]
        Ideas[Self-Contained Ideas]
        CustomComponents[Customizable Components]
    end

    subgraph Core System
        ML[Machine Learning Layer]
        Relations[Relationship Detection]
        Patterns[Pattern Recognition]
    end

    subgraph Context Framework
        Thresh[Context Thresholds]
        Markers[Context Markers]
        Changes[Change Detection]
    end

    subgraph Data Management
        Clean[Clean Data Store]
        Meta[Metadata Processing]
    end

    %% User interactions
    UI --> Ideas
    Ideas --> CustomComponents
    
    %% Core processing
    Ideas --> ML
    ML --> Relations
    ML --> Patterns
    
    %% Context handling
    Relations --> Thresh
    Patterns --> Markers
    Markers --> Changes
    
    %% Data flow
    Changes --> Clean
    Clean --> Meta
    Meta -.-> ML

    %% Component customization
    CustomComponents -.-> Ideas
    
    %% Notes
    classDef default fill:#f9f9f9,stroke:#333,stroke-width:1px
    classDef core fill:#e1f5fe,stroke:#333,stroke-width:1px
    classDef context fill:#e8f5e9,stroke:#333,stroke-width:1px
    classDef data fill:#fff3e0,stroke:#333,stroke-width:1px
    
    class ML,Relations,Patterns core
    class Thresh,Markers,Changes context
    class Clean,Meta data
graph TD subgraph User Layer UI[User Interface] Ideas[Self-Contained Ideas] CustomComponents[Customizable Components] end
subgraph Core System    ML[Machine Learning Layer]    Relations[Relationship Detection]    Patterns[Pattern Recognition]endsubgraph Generative Layer    GenAI[Generative AI Models]    Context[Context Management]    Output[Output Processing]endsubgraph Context Framework    Thresh[Context Thresholds]    Markers[Context Markers]    Changes[Change Detection]endsubgraph Data Management    Clean[Clean Data Store]    Meta[Metadata Processing]end%% User interactionsUI --> IdeasIdeas --> CustomComponents%% Core processingIdeas --> MLML --> RelationsML --> Patterns%% Generative AI flowML --> GenAIRelations --> ContextContext --> GenAIGenAI --> OutputOutput --> UI%% Context handlingRelations --> ThreshPatterns --> MarkersMarkers --> Changes%% Data flowChanges --> CleanClean --> MetaMeta -.-> ML%% Component customizationCustomComponents -.-> Ideas%% NotesclassDef default fill:#f9f9f9,stroke:#333,stroke-width:1pxclassDef core fill:#e1f5fe,stroke:#333,stroke-width:1pxclassDef gen fill:#f3e5f5,stroke:#333,stroke-width:1pxclassDef context fill:#e8f5e9,stroke:#333,stroke-width:1pxclassDef data fill:#fff3e0,stroke:#333,stroke-width:1pxclass ML,Relations,Patterns coreclass GenAI,Context,Output genclass Thresh,Markers,Changes contextclass Clean,Meta data

UNPARTY's true innovation lies in how it wraps generative AI models.

The system:

  1. Uses the ML layer to detect relationships and patterns

  2. Manages context thresholds independently

  3. Feeds clean, relevant context to generative AI models

  4. Processes and presents the output back to users

they're essentially teaching the wrapper how to manage context and relationships for different types of generative AI interactions.

graph TD
    subgraph User Actions [User as Source of Truth]
        Entry[Initial Entry/Notes]
        Reintro[Reintroduction/Exploration]
        Share[Refinement/Sharing]
    end

    subgraph UNPARTY Wrapper
        subgraph Listening Layer
            Listen[Context Listener]
            Order[Order Understanding]
            Trigger[Threshold Detection]
        end

        subgraph Core System
            ML[Machine Learning Layer]
            Relations[Relationship Detection]
            Patterns[Pattern Recognition]
        end

        subgraph Generative Layer
            GenAI[Generative AI Models]
            Context[Context Management]
            Output[Output Processing]
        end

        subgraph Data Management
            Clean[Clean Data Store]
            Meta[Metadata Processing]
        end
    end

    %% User-driven flow
    Entry --> Listen
    Reintro --> Listen
    Share --> Listen
    
    %% Listening layer flow
    Listen --> Order
    Order --> Trigger
    
    %% Conditional activations
    Trigger -- "Threshold Met" --> ML
    ML --> Relations
    Relations --> Context
    Context --> GenAI
    GenAI --> Output
    Output --> Share

    %% Data management
    Relations --> Clean
    Clean --> Meta
    Meta -.-> ML
    
    %% Notes
    classDef user fill:#fce4ec,stroke:#333,stroke-width:2px
    classDef listen fill:#e8f5e9,stroke:#333,stroke-width:1px
    classDef core fill:#e1f5fe,stroke:#333,stroke-width:1px
    classDef gen fill:#f3e5f5,stroke:#333,stroke-width:1px
    classDef data fill:#fff3e0,stroke:#333,stroke-width:1px
    
    class Entry,Reintro,Share user
    class Listen,Order,Trigger listen
    class ML,Relations,Patterns core
    class GenAI,Context,Output gen
    class Clean,Meta data

graph TD
    subgraph User Actions [User as Source of Truth]
        Entry[Initial Entry/Notes]
        Reintro[Reintroduction/Exploration]
        Share[Refinement/Sharing]
    end

    subgraph UNPARTY Wrapper
        subgraph Listening Layer
            Listen[Context Listener]
            Order[Order Understanding]
            Trigger[Threshold Detection]
        end

        subgraph Core System
            ML[Machine Learning Layer]
            Relations[Relationship Detection]
            Patterns[Pattern Recognition]
        end

        subgraph Generative Layer
            GenAI[Generative AI Models]
            Context[Context Management]
            Output[Output Processing]
        end

        subgraph Data Management
            Clean[Clean Data Store]
            Meta[Metadata Processing]
        end
    end

    %% User-driven flow
    Entry --> Listen
    Reintro --> Listen
    Share --> Listen
    
    %% Listening layer flow
    Listen --> Order
    Order --> Trigger
    
    %% Conditional activations
    Trigger -- "Threshold Met" --> ML
    ML --> Relations
    Relations --> Context
    Context --> GenAI
    GenAI --> Output
    Output --> Share

    %% Data management
    Relations --> Clean
    Clean --> Meta
    Meta -.-> ML
    
    %% Notes
    classDef user fill:#fce4ec,stroke:#333,stroke-width:2px
    classDef listen fill:#e8f5e9,stroke:#333,stroke-width:1px
    classDef core fill:#e1f5fe,stroke:#333,stroke-width:1px
    classDef gen fill:#f3e5f5,stroke:#333,stroke-width:1px
    classDef data fill:#fff3e0,stroke:#333,stroke-width:1px
    
    class Entry,Reintro,Share user
    class Listen,Order,Trigger listen
    class ML,Relations,Patterns core
    class GenAI,Context,Output gen
    class Clean,Meta data

UNPARTY can't and shouldn't try to force an idea's development. If an idea isn't progressing through those three stages, it's because the user hasn't naturally taken those steps with it yet.

This is quite different from traditional development approaches where systems might try to:

  • Prompt users to develop ideas further

  • Suggest next steps

  • Evaluate idea quality

  • Push for completion

Instead, UNPARTY simply observes and waits for the user's natural progression. The rule of three isn't a requirement to be met, but rather a pattern that emerges when an idea is ready to become a component.

  • Trust (user remains in control)

  • Privacy (decentralized, user-driven data)

  • Engagement (natural progression rather than forced interaction)

  • Scalability (clean, self-validating components)

graph TD
    subgraph User Actions [User as Source of Truth]
        Entry[Initial Entry/Notes]
        Reintro[Reintroduction/Exploration]
        Share[Refinement/Sharing]
    end
    subgraph UNPARTY Wrapper
        subgraph Listening Layer
            Listen[Context Listener]
            Order[Order Understanding]
            Trigger[Threshold Detection]
        end
        subgraph Core System
            ML[Machine Learning Layer]
            Relations[Relationship Detection]
            Patterns[Pattern Recognition]
        end
        subgraph Generative Layer
            GenAI[Generative AI Models]
            Context[Context Management]
            Output[Output Processing]
        end
        subgraph Data Management
            Clean[Clean Data Store]
            Meta[Metadata Processing]
        end
    end
    %% User-driven flow
    Entry --> Listen
    Reintro --> Listen
    Share --> Listen
    
    %% Listening layer flow
    Listen --> Order
    Order --> Trigger
    
    %% Conditional activations
    Trigger -- "Threshold Met" --> ML
    ML --> Relations
    Relations --> Context
    Context --> GenAI
    GenAI --> Output
    Output --> Share
    %% Data management
    Relations --> Clean
    Clean --> Meta
    Meta -.-> ML
    
    %% Notes
    classDef user fill:#fce4ec,stroke:#333,stroke-width:2px
    classDef listen fill:#e8f5e9,stroke:#333,stroke-width:1px
    classDef core fill:#e1f5fe,stroke:#333,stroke-width:1px
    classDef gen fill:#f3e5f5,stroke:#333,stroke-width:1px
    classDef data fill:#fff3e0,stroke:#333,stroke-width:1px
    
    class Entry,Reintro,Share user
    class Listen,Order,Trigger listen
    class ML,Relations,Patterns core
    class GenAI,Context,Output gen
    class Clean,Meta data

NextKey Architectural Terms

Last updated