AI Engineering Learner
I'm learning AI engineering through hands-on experiments with local models, AI-assisted development, and cloud workflows. My goal is to understand model capabilities and build useful, measurable, and maintainable systems around them.
Explore the portfolio · Visit the Dell Local AI Lab
| Local intelligence | Engineering discipline | Cloud integration |
|---|---|---|
| Exploring what local models can do within real hardware constraints. | Learning to evaluate quality, latency, resource use, and failure modes. | Learning Azure hosting, identity, deployment, and cost-aware design. |
Local inference · Hardware constraints · Hands-on learning
A public project dedicated to running local AI models on a 2017 Dell Precision 7720 workstation. A practical setting for exploring the tradeoffs between model capability and available hardware.
Modern AI on a nine-year-old workstation: the 2017 Dell is the hardware-reuse side of the lab in 2026. The 2024 Alienware adds a newer platform for GPU acceleration and connected-laptop workflows. Together, they represent two complementary approaches to getting more from existing hardware.
2024 hardware · RTX acceleration · Multi-laptop experiments
My newer local AI machine is a 2024 Alienware m16 R2. The focus is making useful AI run on hardware I already own: understanding memory limits, connecting laptops, and learning where custom applications add value.
| This machine | Verified configuration |
|---|---|
| CPU | Intel Core Ultra 9 185H · 16 cores / 22 threads |
| System memory | 32 GB DDR5-5600 · 2 × 16 GB modules |
| Discrete GPU | NVIDIA GeForce RTX 4070 Laptop GPU · 8 GB GDDR6 VRAM |
| Additional processors | Intel Arc integrated graphics and Intel AI Boost NPU |
| Wired experiment interface | Thunderbolt 4 / USB4 via USB-C · rated interface bandwidth up to 40 Gbps |
Installed CPU, RAM, GPU, and USB4/NPU devices checked locally on November 14, 2026. Rated interface capabilities: Dell specifications and owner's manual.
Inside the experiments
- NVIDIA PAIR (Personal AI Router): connecting a peer laptop to an inference engine on the Alienware. Recorded lab tests established cross-laptop request routing, with PAIR selecting an execution node for each request.
- Thunderbolt 4 / USB4 networking: using a direct USB-C cable link for laptop-to-laptop IP communication, model requests, and file transfers. Lab runs measured roughly 26 Gbps network throughput with iperf3 in October 2026.
- Modern local models: experimenting with quantized gpt-oss-20b and Qwen3-30B-A3B models using llama.cpp, with LM Studio and Ollama also explored. GPU offload plus system RAM supports experimentation with larger model configurations.
- CUDA and custom applications: exploring GPU-accelerated application development and practical integrations for local inference.
- Engineering the whole workflow: studying streaming responses, request handling, memory use, and thermal settings alongside generation throughput.
The lab combines completed routing and networking experiments with ongoing model and application exploration. Each experiment builds a clearer picture of how the hardware, runtime, and workflow work together.
Azure · Web development · Learning in public
A developing home for project stories, experiments, and technical notes. The initial portfolio shell is live; detailed write-ups and results will follow as they are ready to share.
- Reproducible experiments: clear inputs, recorded settings, and repeatable comparisons.
- Useful evaluation: checking model answers against evidence and task-specific criteria.
- Retrieval and grounding: learning how to connect model responses to relevant source material.
- Responsible integration: protecting private data and keeping compute and API costs bounded.
My learning approach: build, measure, understand, document
- Start with a question. Define the task and what a useful result would look like.
- Build the smallest experiment. Keep the moving parts understandable.
- Measure and inspect. Look at quality, performance, cost, and the cases that fail.
- Document the tradeoffs. Separate observations from assumptions and revise when the evidence changes.
AI assistants can help develop ideas and code; testing and reviewing their output remain part of the work.
Personal projects. Ongoing learning. Original AI-generated conceptual banner and custom animated workflow illustration.
