{"id":31737532,"url":"https://github.com/jxtngx/deep-learning-with-claude","last_synced_at":"2025-12-30T01:14:02.647Z","repository":{"id":316091127,"uuid":"1061537451","full_name":"jxtngx/deeplearning-with-claude","owner":"jxtngx","description":"A modular, multi-agent based system for PyTorch, Hugging Face, and AWS.","archived":false,"fork":false,"pushed_at":"2025-10-08T10:34:40.000Z","size":320,"stargazers_count":14,"open_issues_count":0,"forks_count":4,"subscribers_count":0,"default_branch":"main","last_synced_at":"2025-10-08T12:27:25.723Z","etag":null,"topics":["artificial-intelligence","claude-code","deep-learning","pytorch"],"latest_commit_sha":null,"homepage":"https://jxtngx.github.io/claude-code-pytorch/","language":"HTML","has_issues":true,"has_wiki":null,"has_pages":null,"mirror_url":null,"source_name":null,"license":"apache-2.0","status":null,"scm":"git","pull_requests_enabled":true,"icon_url":"https://github.com/jxtngx.png","metadata":{"files":{"readme":"README.md","changelog":null,"contributing":null,"funding":null,"license":"LICENSE","code_of_conduct":null,"threat_model":null,"audit":null,"citation":null,"codeowners":null,"security":null,"support":null,"governance":null,"roadmap":null,"authors":null,"dei":null,"publiccode":null,"codemeta":null,"zenodo":null,"notice":null,"maintainers":null,"copyright":null,"agents":null,"dco":null,"cla":null}},"created_at":"2025-09-22T04:00:20.000Z","updated_at":"2025-10-08T10:34:44.000Z","dependencies_parsed_at":null,"dependency_job_id":"fa8bb13f-e330-4f56-94fb-bd08ca43794e","html_url":"https://github.com/jxtngx/deeplearning-with-claude","commit_stats":null,"previous_names":["jxtngx/claude-code-pytorch","jxtngx/deeplearning-with-claude"],"tags_count":0,"template":true,"template_full_name":null,"purl":"pkg:github/jxtngx/deeplearning-with-claude","repository_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/jxtngx%2Fdeeplearning-with-claude","tags_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/jxtngx%2Fdeeplearning-with-claude/tags","releases_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/jxtngx%2Fdeeplearning-with-claude/releases","manifests_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/jxtngx%2Fdeeplearning-with-claude/manifests","owner_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners/jxtngx","download_url":"https://codeload.github.com/jxtngx/deeplearning-with-claude/tar.gz/refs/heads/main","sbom_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/jxtngx%2Fdeeplearning-with-claude/sbom","scorecard":null,"host":{"name":"GitHub","url":"https://github.com","kind":"github","repositories_count":279001126,"owners_count":26083021,"icon_url":"https://github.com/github.png","version":null,"created_at":"2022-05-30T11:31:42.601Z","updated_at":"2022-07-04T15:15:14.044Z","status":"online","status_checked_at":"2025-10-09T02:00:07.460Z","response_time":59,"last_error":null,"robots_txt_status":"success","robots_txt_updated_at":"2025-07-24T06:49:26.215Z","robots_txt_url":"https://github.com/robots.txt","online":true,"can_crawl_api":true,"host_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub","repositories_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories","repository_names_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repository_names","owners_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners"}},"keywords":["artificial-intelligence","claude-code","deep-learning","pytorch"],"created_at":"2025-10-09T09:22:57.939Z","updated_at":"2025-12-30T01:14:02.581Z","avatar_url":"https://github.com/jxtngx.png","language":"HTML","funding_links":[],"categories":[],"sub_categories":[],"readme":"# Deep Learning with Claude\n\n[![Author](https://img.shields.io/badge/author-jxtngx-blue)](https://github.com/jxtngx)\n[![License](https://img.shields.io/badge/License-Apache_2.0-blue.svg)](https://opensource.org/licenses/Apache-2.0)\n\nA modular, multi-agent based system for PyTorch, Hugging Face, and AWS, powered by Anthropic's Claude family of models.\n\n\u003e **Note**: This project is compatible with GitHub Copilot through `.github/copilot-instructions.md`, which references the same agent architecture defined in `CLAUDE.md`.\n\n## Philosophy\n\nThis repository embodies an **agent-based architecture** for machine learning projects, where specialized AI agents collaborate to deliver comprehensive solutions. Each agent maintains deep expertise in their domain while remaining modality and task agnostic.\n\n## Core Principles\n\n\u003ctable\u003e\n\u003ctr\u003e\n\u003ctd style=\"background-color: black; color: #808080; padding: 15px; width: 50%; vertical-align: top;\"\u003e\u003ch3 style=\"margin-top: 0; margin-bottom: 10px; color: white;\"\u003eSeparation of Concerns\u003c/h3\u003eEach agent owns a specific technical domain, preventing overlap and ensuring expertise depth.\u003c/td\u003e\n\u003ctd style=\"background-color: black; color: #808080; padding: 15px; width: 50%; vertical-align: top;\"\u003e\u003ch3 style=\"margin-top: 0; margin-bottom: 10px; color: white;\"\u003eModality Agnostic\u003c/h3\u003eAgents adapt to any ML task—vision, NLP, audio, multimodal—without hardcoded assumptions.\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd style=\"background-color: black; color: #808080; padding: 15px; width: 50%; vertical-align: top;\"\u003e\u003ch3 style=\"margin-top: 0; margin-bottom: 10px; color: white;\"\u003ePerformance First\u003c/h3\u003eOptimized for PyTorch 2.3+ with distributed training and NVIDIA GPU acceleration.\u003c/td\u003e\n\u003ctd style=\"background-color: black; color: #808080; padding: 15px; width: 50%; vertical-align: top;\"\u003e\u003ch3 style=\"margin-top: 0; margin-bottom: 10px; color: white;\"\u003eCloud Native\u003c/h3\u003eBuilt for AWS EC2 environments with scalable infrastructure patterns.\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd colspan=\"2\" align=\"center\" style=\"background-color: black; color: #808080; padding: 15px;\"\u003e\u003ch3 style=\"margin-top: 0; margin-bottom: 10px; color: white;\"\u003eCollaborative Intelligence\u003c/h3\u003eAgents work in concert, sharing context and building on each other's outputs.\u003c/td\u003e\n\u003c/tr\u003e\n\u003c/table\u003e\n\n## Skill Progression Platform\n\nThis template serves as a gateway to two critical ML engineering competencies:\n\n### PyTorch Mastery\nProgress from basic tensor operations to production-ready ML systems through practical, agent-guided development. The `prompting-guide/` provides a structured path from prompt dependency to independent PyTorch expertise.\n\n### Agentic Application Development\nThe multi-agent architecture here provides hands-on experience with patterns directly applicable to:\n- **LangChain**: Chain-of-thought reasoning, tool use, and agent orchestration\n- **AWS Bedrock Agents**: Structured prompts, knowledge bases, and action groups\n- **NVIDIA NeMo Guardrails**: Agent safety, structured outputs, and conversation flows\n\nBy working with this template's agent team, you're learning:\n- Agent coordination patterns (supervisor/worker models)\n- Tool use and function calling (ReAct patterns)\n- Structured prompting (INVEST+CRPG framework)\n- Multi-agent orchestration (parallel and sequential workflows)\n\nThese skills transfer directly to building production agent applications, making this template both a PyTorch learning tool and an introduction to the agentic AI ecosystem.\n\n## Structured Requirements Format\n\nThis project uses a disciplined approach to requirements specification:\n\n### INVEST User Stories\nAll agent tasks and prompt templates follow the agile INVEST criteria\n\n- **Independent** - Each story stands alone\n- **Negotiable** - Flexible implementation details\n- **Valuable** - Clear business or research value\n- **Estimable** - Measurable scope and effort\n- **Small** - Completable in reasonable time\n- **Testable** - Verifiable success criteria\n\n### CRPG Optimization Framework\nA custom format using Reinforcement Learning language guides AI agent optimization\n\n- **Constraints** - Technical boundaries and limitations\n- **Rewards** - Success metrics and performance targets\n- **Penalties** - Anti-patterns and quality deductions\n- **Goal State** - Clear deliverables and validation criteria\n\nThis structured approach ensures agents understand both the \"what\" (user story) and the \"how\" (optimization parameters) of each task.\n\n## Getting Started\n\n1. **Define Your Project**: Consult `CLAUDE.md` to engage the Supervisor\n2. **Select Your Team**: Claude routes to appropriate specialist agents\n3. **Iterate and Build**: Agents collaborate to implement your solution\n\n## Architecture\n\n### Agent Team Structure\n\n```mermaid\ngraph TB\n    %% Strategy Team\n    Supervisor[\"Supervisor - Project Coordination\"]\n    DomainExpert[\"DomainExpert - Domain Knowledge\"]\n\n    %% Data Pipeline Team\n    DatasetCurator[\"DatasetCurator - HF Datasets\"]\n    DataEngineer[\"DataEngineer - DataLoaders\"]\n    TransformSpecialist[\"TransformSpecialist - Augmentation\"]\n\n    %% Model Architecture Team\n    ModelArchitect[\"ModelArchitect - HF Models\"]\n    NetworkArchitect[\"NetworkArchitect - Custom Networks\"]\n\n    %% Training \u0026 Evaluation Team\n    TrainingOrchestrator[\"TrainingOrchestrator - Training Loops\"]\n    MetricsArchitect[\"MetricsArchitect - Evaluation\"]\n    RunnerOrchestrator[\"RunnerOrchestrator - Pipelines\"]\n\n    %% Infrastructure Team\n    CloudEngineer[\"CloudEngineer - AWS Services\"]\n    ComputeOrchestrator[\"ComputeOrchestrator - EC2/GPU\"]\n    LocalStackEmulator[\"LocalStackEmulator - Local Testing\"]\n\n    %% Quality \u0026 Interface Team\n    TestArchitect[\"TestArchitect - TDD\"]\n    InterfaceDesigner[\"InterfaceDesigner - Web UI\"]\n\n    %% Team Groupings\n    subgraph Strategy\n        Supervisor\n        DomainExpert\n    end\n\n    subgraph DataPipeline[Data Pipeline]\n        DatasetCurator\n        DataEngineer\n        TransformSpecialist\n    end\n\n    subgraph ModelArchitecture[Model Architecture]\n        ModelArchitect\n        NetworkArchitect\n    end\n\n    subgraph TrainingEvaluation[Training \u0026 Evaluation]\n        TrainingOrchestrator\n        MetricsArchitect\n        RunnerOrchestrator\n    end\n\n    subgraph Infrastructure\n        CloudEngineer\n        ComputeOrchestrator\n        LocalStackEmulator\n    end\n\n    subgraph QualityInterface[Quality \u0026 Interface]\n        TestArchitect\n        InterfaceDesigner\n    end\n\n    %% Primary Relationships\n    Supervisor --\u003e DatasetCurator\n    Supervisor --\u003e ModelArchitect\n    Supervisor --\u003e CloudEngineer\n\n    DomainExpert --\u003e DatasetCurator\n    DomainExpert --\u003e MetricsArchitect\n\n    DatasetCurator --\u003e DataEngineer\n    DataEngineer --\u003e TransformSpecialist\n    DataEngineer --\u003e TrainingOrchestrator\n\n    ModelArchitect --\u003e NetworkArchitect\n    NetworkArchitect --\u003e TrainingOrchestrator\n\n    TrainingOrchestrator --\u003e MetricsArchitect\n    TrainingOrchestrator --\u003e RunnerOrchestrator\n\n    CloudEngineer --\u003e ComputeOrchestrator\n    CloudEngineer --\u003e InterfaceDesigner\n    LocalStackEmulator --\u003e CloudEngineer\n\n    TestArchitect -.-\u003e DataEngineer\n    TestArchitect -.-\u003e NetworkArchitect\n    TestArchitect -.-\u003e TrainingOrchestrator\n    TestArchitect -.-\u003e CloudEngineer\n\n    RunnerOrchestrator --\u003e ComputeOrchestrator\n\n    %% Styling\n    classDef strategyStyle fill:#e1f5fe,stroke:#0277bd,stroke-width:2px,color:#000\n    classDef dataStyle fill:#f3e5f5,stroke:#6a1b9a,stroke-width:2px,color:#000\n    classDef modelStyle fill:#e8f5e9,stroke:#2e7d32,stroke-width:2px,color:#000\n    classDef trainingStyle fill:#fff3e0,stroke:#e65100,stroke-width:2px,color:#000\n    classDef infraStyle fill:#fce4ec,stroke:#c2185b,stroke-width:2px,color:#000\n    classDef qualityStyle fill:#f1f8e9,stroke:#558b2f,stroke-width:2px,color:#000\n\n    class Supervisor,DomainExpert strategyStyle\n    class DatasetCurator,DataEngineer,TransformSpecialist dataStyle\n    class ModelArchitect,NetworkArchitect modelStyle\n    class TrainingOrchestrator,MetricsArchitect,RunnerOrchestrator trainingStyle\n    class CloudEngineer,ComputeOrchestrator,LocalStackEmulator infraStyle\n    class TestArchitect,InterfaceDesigner qualityStyle\n```\n\n### Workflow\n\n```\nAgents → Specialized Expertise → Collaborative Implementation → Deployed Solution\n```\n\nEach agent operates as an expert consultant, providing:\n- Domain-specific knowledge\n- Best practice implementations\n- Performance optimizations\n- Quality assurance\n\n## Key Technologies\n\n- **PyTorch 2.3+**: Core deep learning framework\n- **Hugging Face**: Model and dataset ecosystem\n- **AWS**: Cloud infrastructure and services\n- **Claude Code**: AI-powered development assistance\n\n## Repository Structure\n\n- `.claude/agents/`: Specialized agent definitions\n- `CLAUDE.md`: Agent routing and coordination\n- `docs/`: Documentation and agile artifacts\n  - `adr/`: Architecture Decision Records\n  - `sprints/`: Sprint planning and tracking\n- `prompt-templates/`: Task-specific prompt examples\n- `prompting-guide/`: Comprehensive guide on prompting techniques and MLE learning path\n- `src/`: Core Python modules (non-package structure)\n  - `data.py`: Data pipeline components\n  - `network.py`: Model architectures\n  - `trainer.py`: Training orchestration\n  - `server.py`: API and serving\n  - `runner.py`: CLI entry point\n\n## Quick Start\n\n### Setup with uv\n\n```bash\n# Install uv (if not already installed)\ncurl -LsSf https://astral.sh/uv/install.sh | sh\n\n# Install dependencies\nuv pip install -r requirements.txt\n\n# Install dev dependencies\nuv pip install -e \".[dev]\"\n\n# Setup pre-commit hooks\npre-commit install\n```\n\n## Using Claude Code Agents\n\n\u003e View all available agents and their capabilities in [.claude/team.md](.claude/team.md)\n\n### Basic Agent Invocation\n\nIn the Claude Code terminal, you can directly invoke specialized agents using `@agent-[NAME]` or let Claude automatically route your request to the appropriate expert.\n\n#### Direct Agent Routing\n```bash\n# Explicitly call a specific agent using @agent-[NAME]\n$ \"@agent-NetworkArchitect implement a custom attention mechanism for video understanding\"\n\n# Agent responds with expertise\nNetworkArchitect: I'll design a custom spatio-temporal attention module...\n```\n\n#### Automatic Routing\n```bash\n# Describe your task and Claude routes to appropriate agents\n$ \"I need to fine-tune a BERT model on my custom dataset with limited GPU memory\"\n\n# Claude automatically engages relevant agents\nSupervisor: Let me establish your constraints...\nTestArchitect: Writing tests for your fine-tuning pipeline...\nModelArchitect: Selecting optimal BERT variant for your memory constraints...\nDataEngineer: Configuring efficient data loading...\n```\n\n### Common Workflows\n\n#### Starting a New Project\n```bash\n$ \"I want to build an image classification system for medical X-rays\"\n\n# Supervisor coordinates the team\nSupervisor: Analyzing requirements...\nDomainExpert: Medical imaging requires specific preprocessing...\nDatasetCurator: Searching for relevant medical datasets...\nTestArchitect: Writing comprehensive test suite first...\nNetworkArchitect: Designing architecture for medical images...\n```\n\n#### Fine-tuning with Limited Resources\n```bash\n$ \"Fine-tune Llama-2-7B on my customer support dataset using QLoRA\"\n\n# Specialized agents collaborate\nModelArchitect: Configuring Llama-2-7B with 4-bit quantization...\nDataEngineer: Setting up efficient data pipeline...\nTrainingOrchestrator: Implementing QLoRA with gradient checkpointing...\nMetricsArchitect: Establishing evaluation metrics...\n```\n\n#### Local Testing Before Deployment\n```bash\n$ \"Test my model API locally before deploying to AWS\"\n\n# LocalStackEmulator coordinates with CloudEngineer\nLocalStackEmulator: Starting local AWS environment...\nCloudEngineer: Configuring API endpoints for local testing...\nTestArchitect: Running integration tests against LocalStack...\n```\n\n#### Creating Test-Driven ML Code\n```bash\n$ \"Write tests for a vision transformer training pipeline\"\n\n# TestArchitect leads TDD workflow\nTestArchitect: Creating tests that will fail initially...\n  - test_model_initialization()\n  - test_forward_pass_shapes()\n  - test_loss_computation()\n  - test_optimizer_step()\nNetworkArchitect: Implementing ViT to pass your tests...\n```\n\n### Multi-Agent Collaboration Example\n\n```bash\n$ \"Deploy a real-time object detection API with \u003c50ms latency\"\n\n# Watch agents collaborate\nSupervisor: Establishing latency requirements...\nTestArchitect: Writing performance benchmarks...\nModelArchitect: Selecting YOLOv8n for speed...\nComputeOrchestrator: Recommending g5.xlarge instance...\nCloudEngineer: Implementing FastAPI with async inference...\nLocalStackEmulator: Testing locally first...\nInterfaceDesigner: Creating monitoring dashboard...\n\n# Result: Complete deployment pipeline with tests\n```\n\n### Tips for Effective Agent Use\n\n1. **Be Specific**: Include constraints, metrics, and requirements\n2. **Direct Invocation**: Use `@agent-[NAME]` to call specific agents\n3. **Use Templates**: Copy prompts from `prompt-templates/` for consistency\n4. **Test First**: Let TestArchitect write tests before implementation\n5. **Local First**: Use LocalStackEmulator before AWS deployment\n6. **Trust Routing**: Claude knows which agents to engage when not specified\n\n### Agent Coordination Patterns\n\n```bash\n# Iterative workflow\n$ \"Test → Data → Model → Training → Deploy (continuous iteration)\"\n\n# Parallel execution\n$ \"Run tests AND start LocalStack AND prepare dataset\"\n\n# Specific expertise request\n$ \"@agent-MetricsArchitect design custom metrics for video quality assessment\"\n```\n\n## Design Principles\n\n### Non-Package Architecture\nThe `src/` directory contains standalone modules that can be run directly without package installation. This simplifies deployment and reduces complexity while maintaining clear separation of concerns.\n\n### Agile Development Process\n- **Architecture Decision Records**: Documented technical decisions in `docs/adr/`\n- **Sprint Tracking**: Comprehensive sprint planning and retrospectives in `docs/sprints/`\n- **Test-Driven Development**: TestArchitect enforces TDD practices\n- **Continuous Integration**: Built into agent collaboration workflows\n\n### Modern Tooling\n- **uv**: Fast, reliable Python package management\n- **Ruff**: Single tool for linting and formatting\n- **Pre-commit**: Automated code quality checks\n- **Type hints**: Full typing support throughout\n\nThis template provides the foundation for any ML project, from research prototypes to production systems.\n\n## Citation\n\nIf you use this project in your research or work, please cite:\n\n```bibtex\n@software{claude_code_pytorch,\n  author = {jxtngx},\n  title = {Claude Code PyTorch: Multi-Agent ML Development Framework},\n  year = {2025},\n  url = {https://github.com/jxtngx/claude-code-pytorch},\n  license = {Apache-2.0}\n}\n```\n\n## License\n\nCopyright 2025 jxtngx\n\nLicensed under the Apache License, Version 2.0. See [LICENSE](LICENSE) for details.\n","project_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fjxtngx%2Fdeep-learning-with-claude","html_url":"https://awesome.ecosyste.ms/projects/github.com%2Fjxtngx%2Fdeep-learning-with-claude","lists_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fjxtngx%2Fdeep-learning-with-claude/lists"}