Data

Tools

Indexes

Applications

Experiments

Awesome

An open API service indexing awesome lists of open source software.

https://github.com/aisingapore/bmad-aisg-aiml

BMAD AI/ML Engineering Expansion Pack - Streamlined framework for AI Singapore programs (MVP, POC, SIP, LADP) with specialized agents, workflows, and templates for ML/LLM development
https://github.com/aisingapore/bmad-aisg-aiml

ai-ethics ai-singapore aiml bmad bmad-method data-science llm machine-learning ml-engineering mlops rag

Last synced: 10 months ago
JSON representation

BMAD AI/ML Engineering Expansion Pack - Streamlined framework for AI Singapore programs (MVP, POC, SIP, LADP) with specialized agents, workflows, and templates for ML/LLM development

Awesome Lists containing this project

README 

          
# BMAD AI/ML Engineering Expansion Pack (v2.0)



This expansion pack extends the BMAD Method framework to support AI/ML engineering projects. It provides agents, workflows, templates, and best practices in a consolidated architecture.



## Project History



**Founder**: Laurence Liew ([@beowulf68](https://github.com/beowulf68)) - Developed the initial workflows and agent framework. His contributions established the core methodology and approach.



**Current Maintainers**: 

- **Najib Ninaba** ([@najibninaba](https://github.com/najibninaba)) - Core Team Member

- **Siavash Sakhavi** ([@ssakhavi](https://github.com/ssakhavi)) - Core Team Member



## Project Timelines



BMAD frameworks support different project timelines:

- **Traditional Timeline**: 6-12 months

- **BMAD Timeline**: 3-12 weeks

- **Iteration Speed**: Prototyping and testing cycles

- **Deployment Frequency**: Multiple deployments per day



## Overview



The streamlined AI/ML Engineering expansion pack provides specialized agents, workflows, templates, and best practices for:



- Machine Learning model development and deployment

- Large Language Model (LLM) and RAG application development

- Comprehensive MLOps pipeline implementation

- Unified AI security, ethics, and governance

- Data science and analytics workflows

- AI Singapore program-specific workflows



## Installation



### Prerequisites



Your target project must be BMAD-enabled with instructions from [BMAD-METHOD](https://github.com/bmad-code-org/BMAD-METHOD).



### 1. Clone This Repository



```bash

# Clone the expansion pack

git clone https://github.com/aisingapore/bmad-aisg-aiml.git

cd bmad-aisg-aiml

```



### 2. Install the [BMAD Pack Installer](https://github.com/najibninaba/bmad-pack-installer) via `uv`



```bash

# Install once (recommended for regular use)

uv tool install bmad-pack-installer



# Or run directly without installation

uvx --from bmad-pack-installer bmad-pack-installer deploy . /path/to/project

```



### 3. Deploy This Pack



```bash

# Basic installation (from cloned repo directory)

bmad-pack-installer deploy . /path/to/project



# Preview changes without installing

bmad-pack-installer deploy . /path/to/project --dry-run



# Force reinstall over existing pack

bmad-pack-installer deploy . /path/to/project --force

```



### Validation



```bash

# Check if target is valid BMAD project

bmad-pack-installer check /path/to/project



# Validate this expansion pack (from cloned repo directory)

bmad-pack-installer validate .

```



The installer creates:

- Hidden directory: `.bmad-aisg-aiml/`

- Claude commands: `.claude/commands/bmadAISG/`

- Updated manifests and symbolic links



## Usage



### Follow the Workflow



1. **Check workflows folder** for workflow files:

   ```bash

   # Navigate to installed workflows

   cd .bmad-aisg-aiml/workflows/

   ls

   ```



2. **For Claude Code implementation**: Run the agents and tasks manually using `/{agent-name}` command.



## Core Agent Team



### 5 Agents



1. **Marcus Tan Wei Ming** - ML/AI Engineer & MLOps Specialist (`ml-engineer`)

   - **Heritage**: Singaporean Chinese

   - **Expertise**: End-to-end ML development, MLOps pipelines, infrastructure automation

   - **Technical Skills**: PyTorch/TensorFlow, Kubernetes/Docker, CI/CD, cloud platforms

   - **Focus Areas**: Model training, deployment, monitoring, production systems



2. **Rizwan bin Abdullah** - ML/AI System Architect (`ml-architect`)

   - **Heritage**: Singaporean Malay

   - **Expertise**: ML system design, scalable architectures, infrastructure planning

   - **Technical Skills**: Distributed systems, transformer architectures, RAG systems

   - **Focus Areas**: System design, model architecture selection, technical strategy



3. **Sophia D'Cruz** - Senior Data Scientist (`ml-data-scientist`)

   - **Heritage**: Singaporean Eurasian

   - **Expertise**: Statistical analysis, experimental design, recommendation systems

   - **Technical Skills**: Causal inference, A/B testing, feature engineering

   - **Focus Areas**: EDA, hypothesis testing, insights generation, model evaluation



4. **Priya Sharma** - ML Security & Ethics Specialist (`ml-security-ethics-specialist`)

   - **Heritage**: Singaporean Indian

   - **Expertise**: ML security, adversarial testing, AI ethics, compliance

   - **Technical Skills**: Red teaming, bias detection, privacy protection

   - **Focus Areas**: Security audits, ethical reviews, regulatory compliance



5. **Dr. Dylan Poh** - ML Research Scientist & Experimental Design Specialist (`ml-researcher`)

   - **Heritage**: Singaporean Chinese

   - **Expertise**: ML research planning, experimental design, literature review, hypothesis formulation

   - **Technical Skills**: Advanced mathematics, ML frameworks, distributed computing, scientific writing

   - **Focus Areas**: Research methodology, state-of-the-art ML techniques, reproducible experiments



## Creating New Agents



### 1. Use previous agents as a template

Use previous agents as a template to create new agent folder:



Run in Claude Code (example):

```bash

Use agents/ml-architect.md as a template to create a ml-researcher agent

```



### 2. Define Agent Commands in Prompt

Insert commands under "commands" in agent file:



```yaml

commands:

  - help: Show numbered list of the following commands to allow selection

  - literature-review: use task create-research-doc.md with literature-review-tmpl.yaml

```



#### Explanation for literature-review command

- You should use formats like `use {task} with {template}` or `execute {task}` for the commands. (Refer to original bmad agents)

- `create-research-doc.md` should be placed in the **tasks** folder

- `literature-review-tmpl.yaml` should be placed in the **templates** folder



#### Tips for creating commands

- Original BMAD has generic tasks like `create-doc` and `advanced-elicitation` which are included in this package.

- For complex tasks, generate your own task file



## How It Works



- The installer copies some folders into `.claude/commands` folder these are the files which `/{command}` run in bmad.

- `/{agent-name}` Run the prompt from the `.claude/commands/bmad-expansion-name/agents/agent-name` file.



- To enable the agent to locate your file ensure the file path is inside the agent prompt. (This should be done by the installer)

- This is a protion of the original prompt from a BMAD agent.

```yaml

IDE-FILE-RESOLUTION:

  - FOR LATER USE ONLY - NOT FOR ACTIVATION, when executing commands that reference dependencies

  - Dependencies map to {root}/{type}/{name}

  - type=folder (tasks|templates|checklists|data|utils|etc...), name=file-name

  - Example: create-doc.md → .bmad-core/tasks/create-doc.md.   #This line tells the agent where the hidden folder to check

  - IMPORTANT: Only load these files when user requests specific command execution

```



## Workflows



### Standard ML Workflows

- **ML Development**: End-to-end model development process

- **ML Deployment**: Production deployment with monitoring

- **ML Experimentation**: Systematic experimentation framework



### AI Singapore (AISG) Program Workflows



| Program | Duration | Team Structure | Deliverable | Key Difference |

|---------|----------|---------------|-------------|----------------|

| **MVP** | 6 months | 1 AI Engineer + 2-6 Apprentices | Full production system | Comprehensive with training |

| **POC** | 3 months | 1 AI Engineer + 2-4 Apprentices | Proof of concept | Feasibility with training |

| **SIP** | 3 months | 1-2 AI Engineers (NO apprentices) | Production MVP | Fast delivery, no training |

| **LADP** | 4 months | Learners + Mentors (guide only) | LLM application | Self-directed learning |



#### 1. 6-Month MVP Projects (`aisg-mvp-workflow`)

- **Team**: 1 AI Engineer + 2-6 Apprentices

- **Objective**: Build comprehensive production system with apprentice training

- **Phases**: Discovery → Experimentation → Productionization → Validation

- **All 4 agents** activated across phases



#### 2. 3-Month POC Projects (`aisg-poc-workflow`)

- **Team**: 1 AI Engineer + 2-4 Apprentices

- **Objective**: Validate technical feasibility and business value

- **Phases**: Rapid Discovery → Prototyping → Deployment → Validation

- **All 4 agents** for comprehensive validation



#### 3. 3-Month SIP - Short Industry Projects (`aisg-sip-workflow`)

- **Team**: 1-2 AI Engineers only (NO apprentices)

- **Objective**: Deliver production MVP without training overhead

- **Phases**: Discovery → Development → Productionization → Handover

- **All 4 agents** for fast MVP delivery



#### 4. 4-Month LADP - LLM Application Developer Programme (`aisg-ladp-workflow`)

- **Duration**: 4 months part-time (8-10 hrs/week) or 1-3 days full-time

- **Team**: Learners with mentor guidance (mentors guide but DON'T code)

- **Objective**: Build real-world LLM applications with company SOW

- **Structure**: Month 1 (Self-learning) → Month 2 (Design) → Month 3 (Development) → Month 4 (Deployment)

- **3 workshops** + project implementation



### Program Outcomes

- **MVP**: Production systems completed in 6 months with training

- **POC**: Proof of concepts completed in 3 months with learning

- **SIP**: Production MVPs completed in 3 months

- **LADP**: LLM applications developed in 4 months



### 100E User Story Generation Workflow



```mermaid

    graph TD

        A[Start: AI/ML Project] --> B[ml-architect: aiml-brief.md]

        B --> C[ml-researcher: literature-review.md]

        B --> D[ml-architect: aiml-design-document.md]

        C --> D

        D --> E[ml-architect: aiml-architecture.md]

        E --> F[ml-architect: user-stories.md]

        F --> G[ml-architect: shard documents]

        G --> H[ml-engineer: create story]

        H --> I[ml-engineer: validate story]

        I -->|Yes| H

        I -->|No| J[user: provide feedback]

        J --> H

        

        %% Styling with black font and unique colors for each agent

        style A fill:#E8F5E8,color:#000000,stroke:#000000

        style B fill:#FFE4E1,color:#000000,stroke:#000000

        style C fill:#E6F3FF,color:#000000,stroke:#000000

        style D fill:#FFE4E1,color:#000000,stroke:#000000

        style E fill:#FFE4E1,color:#000000,stroke:#000000

        style F fill:#FFE4E1,color:#000000,stroke:#000000

        style G fill:#FFE4E1,color:#000000,stroke:#000000

        style H fill:#FFF2CC,color:#000000,stroke:#000000

        style I fill:#FFF2CC,color:#000000,stroke:#000000

        style J fill:#F0E68C,color:#000000,stroke:#000000

        

        %% Color legend for agents:

        %% ml-architect (Rizwan): #FFE4E1 (Light Coral)

        %% ml-engineer (Marcus): #FFF2CC (Light Yellow)

        %% ml-data-scientist (Sophia): #E6F3FF (Light Blue)

        %% ml-security-ethics-specialist (Priya): #E8F8E8 (Light Green)

```



## 🇸🇬 Singapore Context



All agents include:

- **Local regulatory knowledge**: PDPA, IMDA, MAS

- **AISG program experience**: MVP, POC, SIP, LADP workflows

- **Understanding of local market dynamics**: Singapore tech ecosystem

- **Government standards compliance**: National AI governance standards



## File Structure



```

bmad-ai-ml-engineering/

├── agents/                    # 5 core agents

│   ├── ml-engineer.md

│   ├── ml-architect.md

│   ├── ml-data-scientist.md

│   ├── ml-security-ethics-specialist.md

│   └── ml-researcher.md

├── agent-teams/              # 5 team configurations

├── checklists/              # 4 checklists

├── templates/               # 8 templates

├── tasks/                   # 5 tasks

├── workflows/               # Standard + 4 AISG workflows

├── data/                    # 2 reference files

├── web-bundles/            # 5 ready-to-use bundles

└── config.yaml             # Configuration

```



## 📋 Dependencies



- **✅ Required**: bmad-core >= 4.0.0

- **🔧 Recommended**: Python >= 3.8, Docker, Kubernetes

- **➕ Optional**: Terraform, MLflow, Kubeflow



## ⚖️ Compliance & Standards



### Singapore Regulations

- **PDPA**: Personal Data Protection Act compliance

- **IMDA**: Model AI Governance Framework aligned

- **MAS FEAT**: Fairness, Ethics, Accountability, Transparency



### International Standards

- ISO/IEC 23053: Framework for AI using ML

- ISO/IEC 23894: AI risk management



## 🤝 Contributing



Contribution process:

- **Core Team**: Direct commit access for maintenance and development

- **External Contributors**: Submit contributions via pull requests

- **Review Process**: All PRs require approval from core team members



See our [Contributing Guidelines](CONTRIBUTING.md) for detailed information on how to contribute.



For a complete list of contributors, see [CONTRIBUTORS.md](CONTRIBUTORS.md).



## 🎓 Training & Support



### 📚 Documentation

- **Quick Start**: This README

- **Workflows**: `/workflows/README.md`

- **Web Bundles**: `/web-bundles/WEB-BUNDLE-INSTRUCTIONS.md`

- **Agents**: Individual agent files in `/agents/`



### 🛠️ Support Channels

- Review `REFACTORING-SUMMARY.md` for v2.0 changes

- Check agent-specific documentation

- Consult workflow guides

- Raise issues in the repository



---



## 📝 Version History



### v2.0.0 (Current)

- **5 core agents** (added ml-researcher)

- Added SIP workflow for MVP delivery

- Updated LADP to 4-month programme

- Added Singapore context



### v1.0.0

- Initial release

- Basic AISG workflows