Mobile Solutions Architect & Senior Software Engineer with 15+ years of experience designing scalable enterprise mobility tools, automation platforms, and LLM-based assistant systems.

I specialize in:

• 🔐 Mobile Security (MDM/MAM – Intune, Workspace ONE, Knox)
• 📱 Native iOS & Android development
• 🔄 Enterprise automation (API integration, workflows, device diagnostics)
• 🧠 LLM-based tooling (QLoRA, LangChain, custom vector pipelines)

⚠️ Actively publishing refactored tools from my enterprise portfolio. Follow along as I surface, modernize, and document years of technical work.

E.M.A. is a private research project focused on building scalable, AI-driven automation tools for enterprise platforms that exposes REST API's. It combines distributed processing, LLMs, and custom pipelines to create workflows for enterprise data orchestration, QA generation, and model training.

⚠️ This project is under active development and currently private.

• QA Pair Generation:
  • Generates high-quality question-answer pairs for domain-specific datasets.
  • Supports multiple question types: fact-based, why/how, multiple-choice, fill-in-the-blank, and true/false.
• ChromaDB Integration:
  • Enables retrieval-augmented generation (RAG) by storing and retrieving embeddings.
• Fine-Tuning Ready:
  • Supports fine-tuning large language models (e.g., LLaMA) with QLoRA for domain-specific tasks.
• Distributed Processing:
  • Scalable workload distribution using Ray for parallel processing.
• Validation and Cleaning:
  • Ensures generated QA pairs meet quality and consistency standards.
• Centralized Configuration:
  • YAML files for managing global and component-specific settings.
• Expansion Layer:
  • Multi-model QA generation and NER workflows.
  • Directed prompt engineering logic for multi-purpose dataset creation.
• Preprocessing Layer:
  • Includes sliding-window context building, dataset cleaning, and key splitting.
• Distributed Layer:
  • Ray-based JSONL cleanup, sharding, and batch dispatching to models.
• Fine-Tuning Layer:
  • Deepspeed pipeline trainer, LoRA config support, sliding context batching.

E.M.A/
├── backend/                      # FastAPI backend
├── configs/                     # YAML config files
├── data_layer/                  # ChromaDB & ingestion logic
├── distributed_data_layer/     # Large JSONL datasets for distributed QA
├── distributed_processing/     # Ray actors and file distribution logic
├── embedding/                  # Embedding generation scripts
├── expansion_layer/            # Multi-model QA/NER workflows
├── fine_tuning_layer/          # Deepspeed, QLoRA, dataset processing
├── frontend/                   # React-based interface
├── ingestion_layer/            # API/KB/Docs scraping utilities
├── md_notes/                   # Markdown project notes
├── model_layer/                # LLM engine & tool coordination logic
├── preprocessing_layer/       # JSONL deduplication & formatting
├── qa_generation/             # Legacy QA generation logic
├── ray_cluster/               # Ray head/worker node management
├── scripts/                   # Orchestration and utility scripts
├── utilities/                 # Logging, validation, system tools
├── workspace_one_workflows/  # Workspace ONE automation flows
└── .env / requirements.txt    # Environment & dependencies

1. Distributed Dataset Prep:

   • Shard, deduplicate, and clean raw datasets with distributed_cleaning_pipeline.py.

2. Dataset Preparation:

   • Validate and clean input datasets (e.g., omnissa_apis_with_context_dataset.jsonl).
   • Generate embeddings and ingest them into ChromaDB.

3. QA Pair Generation:

   • Run the QA generation pipeline to create question-answer pairs using run_qa_generation.py.

4. Fine-Tuning:

   • Use run_fine_tuning.py to launch the deepspeed_trainer.py or pipeline_trainer.py.
   • Supports QLoRA, DeepSpeed ZeRO-3, sliding context batching, and Flash Attention.

5. Deployment:

   • Deploy the fine-tuned model with the FastAPI backend for real-time inference with token streaming.

• Support for additional question types.
• Automated hyperparameter tuning for fine-tuning workflows.
• Advanced monitoring and analytics for QA pair generation.
• LoRA weight merging utilities
• Fine-tuned model hub export and quantization
• Dataset tokenizer profiling for context budget planning

A fully native Android WebView wrapper for USB-connected Epson receipt printers using the ePOS2 SDK. Built for Android Enterprise deployments with a clean MVVM architecture and structured JSON print support.

A dynamic staging tool designed to configure and provision devices via MDM assignment groups, tag logic, and relay APIs.

Python-based automation framework for interacting with Workspace ONE UEM APIs, featuring DTO mapping and REST abstraction.

• 📦 WorkspaceONE-To-Intune-iOS
  Seamless COPE/BYOD migration utility for iOS MDM transitions.

• 🔬 EasyRest
  Lightweight REST client for debugging iOS APIs.

• 💬 XMPPMessenger-iOS
  Secure real-time chat app built on XMPP.

• 💍 The Proposal
  SpriteKit game with a surprise engagement ending.

• 🧪 IPCDeviceUtility
  Internal sled diagnostic tool with MSR/scanner/firmware support.

• Mobile: Swift, Objective-C, Kotlin, Java
• Frontend: React Typescript Next.js Tailwind CSS PHP
• Backend: Python (FastAPI, Flask), Node.js, Java (SpringBoot), C++, C# .NET, PHP
• DevOps: GitHub Actions, CI/CD, scripting, Jenkins, Sonarqube
• AI/LLM: LangChain, QLoRA, vector DBs, agent frameworks
• Platform Agnostic

• 🌐 noblesite.net
• 💼 LinkedIn
• 📧 noblesite [at] gmail [dot] com

“Build fast. Stay secure. Leave tech cleaner than you found it.”