WisLLM

LLM Agentic Development Team, designed to support business-as-code platforms, services, and data.

MVP Goals

Develop comprehensive requirements documentation for a project.
Copilot instructons for each LLM/Agent should be generated from the requirements.
Write the initial stories necesary to implement the project.
Develop a plan and priorities to tackle the project in an agile manner.
Scrum Master LLM should be able to delegate tasks to the appropriate agents.
Each LLM developer should focus on their specific expertise and delivery only on the next iterative task

Design Philosophy

Locally Capable: Each agent and LLM should be capable of running on local computers.
Energy Efficiency: The system should be designed to minimize energy consumption, allowing for efficient operation on local hardware.
Feedback Required: Every agent should demand and require feedback from one another or the user.
Agentic: Agents should be able to operate independently, but also collaborate when necessary.
Collaborative: Agents should be able to work together to achieve a common goal.
Agile Development: The system should be designed to support agile development practices, allowing for rapid iteration and improvement.
Complexity: LLM Context use-cases and windows should be minimized to reduce complexity and improve performance.
Modular: The system should be modular, allowing for easy integration or swapping of new agents and services.

WisLLM Design Architecture

flowchart TD
    subgraph Users
        User[Business Users]
    end

    subgraph Design[Design LLMs]
        Scrum[Scrum Master]
        Architect[Solutions Architect]
        DataScientist[Data Scientist]
        Researcher[Requirements Researcher]
    end

    subgraph Collab[Collaboration Protocols]
        LangChain[LangChain State Management]
    end

    subgraph Dev[Software Development LLMs]
        UXUI[UX/UI Engineer]
        Backend[Backend Engineer]
        Data[Data Engineer]
    end

    subgraph Ops[Operations LLMs]
        DevSecOps[DevSecOps Engineer]
        Infrastructure[Automation Engineer]
        Reporter[Reporting Analyst]
    end

    subgraph Int[Interface Protocols]
        WisLLM[WisLLM API]
    end

    User <--> Scrum
    User <--> Architect
    User <--> DataScientist

    Architect <--> Scrum
    Architect <--> DataScientist
    
    Architect <--> Researcher
    DataScientist <--> Scrum
    DataScientist <--> Researcher

    Design <--> Collab
    Collab <--> Dev
    Collab <--> Ops

    Dev --> Int
    Ops --> Int

Target Platform LLMs

These LLMs are currently targeted for MVP deployment, and are subject to change based on evolving projects needs, technology advancements, and team feedback.

Reasoning & Architecture Agents: Manages the project, prioritizes tasks, and ensures smooth collaboration between agents.
- NVIDIA Nemotron-H family. The 2/4.8/9bn hybrid Mamba-Transformer models achieve instruction following and code-generation accuracy comparable to dense 30bn LLMs of the same generation at an order-of-magnitude fraction of the inference FLOPs.
- Microsoft Phi series.
  - https://huggingface.co/microsoft/Phi-3.5-mini-instruct - Solutions Architect/Scrum Master
  - https://huggingface.co/microsoft/Phi-4-mini-reasoning - Data Scientist
Code & Data Development Agents: Develops and maintains the project codebase, ensuring high-quality software development practices.
- Falcon-3-10B-1.58bit-Instruct - https://huggingface.co/tiiuae/Falcon3-10B-Instruct-1.58bit
Operations Engineering Agents:
- Salesforce xLAM-2-8B. The 8bn model achieves state-of-the-art performance on tool calling despite is relatively modest size, surpassing frontier models like GPT-4o and Claude 3.5.

Research References

https://arxiv.org/pdf/2506.02153 - Small Language Models are the Future of Agentic AI
https://arxiv.org/pdf/2402.17764 -
https://github.com/microsoft/BitNet/blob/main/gpu/README.md - BitNet GPU Inference
https://arxiv.org/abs/2504.12285 - BitNet Technical Report
https://llm-stats.com/

Folder Structure

WisLLM/
├── pyproject.toml                     # Root workspace configuration
├── scripts/
│   ├── build.sh                       # Build script for the project
│   ├── deploy.sh                      # Deployment script
│   └── run-local.sh                   # Run script for the project
├── src/      
│   ├── __init__.py                    # Package initialization
│   ├── main.py
│   ├── agents/                        # Main entry point for the applicatio
│   |   ├── scrum_master/              # Scrum Master Agent
│   |   │   ├── langgraph.json
│   |   |   ├── graph.py
│   |   |   ├── nodes.py
│   |   |   └── prompts.py
│   |   ├── researcher/                # Requirements Researcher Agent
│   |   ├── data_engineer/             # Data Engineer Agent
│   |   ├── data_scientist/            # Data Scientist Agent
│   |   ├── uiux_engineer/             # UX/UI Engineer Agent
│   |   ├── devsecops_engineer/        # DevSecOps Engineer Agent
│   |   ├── automation_engineer/       # Automation Engineer Agent
│   |   ├── architect/                 # Solutions Architect Agent
│   |   ├── report_analyst/
│   |   └── notification_agent/
│   └── shared/                        # Shared libraries for all agents
│       ├── state/                     # State management
│       │   ├── schemas.py             # Pydantic schemas
│       │   ├── store.py               # State store interface
│       │   └── sync.py                # Synchronization
│       ├── tools/                     # Common tools used by the agent
│       │   ├── cli/                   # Command-line interface tools
│       │   ├── chat/                  # Chat interface
│       │   └── networking/            # Network interface (services/integration?)
│       ├── agents/                    # Base agent classes
│       │   └── base_agent.py
│       └── comms/                     # Inter-agent communication
│           └── handoffs.py
│
└── infrastructure/
    ├── docker/
    └── kubernetes/