Software engineer. Builder. Tinkerer.

I build AWS serverless products, created two brands, and now run a local LLM homelab that pushes open models as fast as the hardware allows.

GitHub Resume Email

Work

Software & AI projects

Production systems built end to end, from API and data model to auth and frontend.

Operations CRM

Unifies Outlook, Amazon / eBay / Shopify sales, tickets, finances, and an AI assistant with memory and RAG on a serverless AWS backend.

Next.jsAmplifyAPI GatewayLambdaDynamoDBCognitoBedrock

Saikospeed Communities

A full-stack community for car enthusiasts: posts, comments, follows, media uploads, and real-time interactions.

AmplifyCognitoDynamoDBLambdaRealtimeS3

Native GPT

Local-first, model-agnostic chat: a Tauri desktop shell with an embedded server, reachable from phones as a PWA over Tailscale, with a Strands Agents loop.

TauriRustReactPythonStrands SDKPWA

Open Code Review Manager

Local-first control plane for Alibaba OpenCodeReview: durable review queue, live progress, structured findings, usage analytics, and an MCP server for agents.

FastAPISQLAlchemyReactTypeScriptMCP

Local LLM

Homelab & open source

Four DGX Sparks and a four-GPU RTX PRO 6000 workstation, plus the tools, kernel work, and hardware I publish while running open models locally.

Inference cluster

4× NVIDIA DGX Spark

Always-on GB10 cluster for tensor-parallel serving, multi-node experiments, and 24/7 local inference.

• Unified memory512 GB (4 × 128 GB)
• InterconnectConnectX-7 200 GbE
• AccessDocker, Tailscale

Read the cluster write-up

Workstation

4× RTX PRO 6000 Blackwell

Threadripper PRO host for FP4 serving, kernel work, and benchmarking from one to four GPUs.

• VRAM384 GB (4 × 96 GB)
• CPUAMD Threadripper PRO
• System RAM256 GB

Cooling

Custom fan shrouds

3D-printed shrouds with an RP2040 closed-loop fan controller, so long runs don't throttle.

• Peak dropUp to 18.6 °C
• ControlRP2040 + host app
• Fits6 GB10 chassis

Software: vLLM, SGLang, llama.cpp, NVFP4, Speculative decoding, Triton, CUDA graphs, Docker, Tailscale.

Open-source work

• SparkDeck Local-first control plane for GB10 clusters: deploy, monitor, and benchmark vLLM, SGLang, and llama.cpp across every node. Python · vLLM · SGLang · llama.cpp 300+merged PRs
• Qwen3.8 Flash-Next FP4 · 1× RTX PRO 6000 Single-GPU inference work: pruned and dynamic MTP draft vocabulary, depth-4 speculation, and exact CUDA-graph sizes. Triton · MTP · NVFP4 +11.5%C1 coding tok/s
• DGX Spark · Qwen3.8-27B NVFP4 Fix and replication guide for tool calling on NVIDIA's vLLM 26.07 image (xgrammar mismatch), with MTP speculative decoding. Shell · vLLM · Tool calling ~79tok/s on one Spark
• DGX Spark Fan Shroud Open hardware: STL/STEP files, schematics, RP2040 firmware, and a host app for temperature-driven fan control on GB10 systems. Hardware · RP2040 · CAD −18 °CGPU peak

All Local LLM projects

Brands

Brands I built

Motorsport ecommerce with an engineering mindset: Shopify, custom integrations, analytics, and automated operations.

Saikospeed

Performance-driven products with a bias toward clean design and measurable results.

ShopifyCustom appsAnalyticsOps automation

Happy Racer Development

Engineering-focused storytelling: build trust through details, not hype.

ShopifyIntegrationsConversionContent ops

Tools

Small utilities

Calculators I built for my own decisions and kept public.

• E85 cost calculatorE85 vs pump gas, cost per mile
• Rent vs ownMortgage, tax, HOA, and invested savings over time

Collection

Blue chip NFTs

A few pieces I hold. Use the arrows or your keyboard to browse.

Contact

Get in touch

Email is the fastest way to reach me: mark95chang@gmail.com.