Skip to main content
RSS
>Michael Svendsen // Senior Backend & AI Platform Engineer

Michael Svendsen — Senior Backend & AI Platform Engineer

Welcome to the official personal website and systems engineering publication of Michael Svendsen (thesven). Based in Aurora / Toronto, Canada, Michael specializes in high-throughput data lakes, Snowflake, dbt, AWS Bedrock AgentCore multi-agent systems, and open-source Model Context Protocol (MCP) tooling.

Featured Technical Publications

  • Deconstructing the October 2026 Shopify Checkout Hijack: Client-Side Supply Chains, AI-Assisted Skimming, and Platform Hardening

    In October 2026, an ecosystem-wide checkout hijacking campaign compromised hundreds of Shopify storefronts through third-party app configurations. Shopify's core platform and native checkout were never breached. Instead, attackers weaponized client-side supply chains, evading CSP and synthetic scanners to execute double-entry skimming on look-alike .top domains. Here is an exhaustive, staff-level architectural teardown of the exploit mechanics, the MITRE ATT&CK vectors, the 563-app delisting wave, and the permanent security countermeasures Shopify has deployed—including ScriptTag deprecation, AST security checks, expiring offline access tokens, and API IP allowlisting.

    Category: Security & Architecture | Tags: #security, #shopify, #app-security, #ecommerce, #javascript, #supply-chain, #architecture | Read Time: 16 min read

  • Architecting a Resilient E-Commerce Analytics Engine: Snowflake, dbt & The Single Source of Truth for Autonomous Agents

    When building high-volume e-commerce platforms, bolting AI agents onto disparate data stores creates fatal metric dissonance. Here is how to engineer a production-grade analytics foundation using Snowflake and dbt that cuts query times by 50% while serving as the authoritative single source of truth for both merchant dashboards and autonomous AI agents.

    Category: Data Platform | Tags: #data-engineering, #snowflake, #dbt, #ai-agents, #shopify, #sql | Read Time: 12 min read

  • SPYPOINT MCP: Bridging Cellular Trail-Camera Telemetry with AI Vision Agents

    Trail cameras capture thousands of photos, but turning raw cellular photo streams into actionable field intelligence requires more than simple image downloads. Here is how I engineered an open-source Model Context Protocol (MCP) server for SPYPOINT cellular cameras—featuring burst clustering algorithms that group 146 photos into 14 distinct visits, strict read-only safety guardrails, and automated contact sheet generation.

    Category: Applied AI & IoT | Tags: #mcp, #python, #iot, #wildlife-photography, #ai, #testing | Read Time: 11 min read

  • Architecting LiteMCP: Bringing Model Context Protocol to Cloudflare Workers

    Anthropic's Model Context Protocol (MCP) revolutionizes how LLMs interact with external tools and data, but official SDKs rely heavily on Node.js internals like stdio, net, and child_process. Here is how I architected LiteMCP—a high-performance, edge-native TypeScript framework designed specifically for Cloudflare Workers with zero cold starts, Web Standards SSE streaming, and full JSON-RPC 2.0 compliance.

    Category: Edge AI & Systems | Tags: #ai, #mcp, #cloudflare, #typescript, #edge-computing, #serverless | Read Time: 11 min read

  • Multi-Agent Orchestration with AWS Bedrock AgentCore: Autonomous Storefront Optimization

    Most enterprise AI features remain bolted-on chatbots generating dead text that cannot mutate state. In this architectural deep dive, we examine how to orchestrate autonomous multi-agent systems using Amazon Bedrock AgentCore and CopilotKit for a high-volume mobile e-commerce platform—turning high-level merchant directives into live, verified storefront layouts with zero server management.

    Category: Agentic Architecture | Tags: #aws, #bedrock, #agentcore, #copilotkit, #multi-agent, #typescript, #architecture | Read Time: 13 min read

  • Automating GGUF Quantization: Bridging Hugging Face Transformers to Local Runtimes

    While open-source inference engines like llama.cpp and Ollama have democratized local LLMs, converting raw Hugging Face checkpoints into production-grade GGUF binaries remains a minefield of broken tokenizers, out-of-memory crashes on sharded safetensors, and ad-hoc quantization flags. Here is how I built GGUF-n-Go to fully automate end-to-end model quantization and multi-precision matrix deployment.

    Category: Machine Learning | Tags: #llm, #gguf, #quantization, #huggingface, #python, #machine-learning, #performance | Read Time: 12 min read

  • Designing a High-Frequency Motorsport Telemetry SDK: Reverse-Engineering & Type-Safe Architecture in TypeScript

    Sim racing telemetry powers real-time driver coaching, team endurance race strategy, and championship analytics. But iRacing's REST API is notorious for aggressive rate limits, asynchronous two-step S3 pre-signed downloads, and shifting JSON schemas. Here is how I architected iracing-data-sdk in TypeScript to provide bulletproof token-bucket throttling, rolling session management, and end-to-end type safety.

    Category: Developer Tooling | Tags: #typescript, #sdk, #telemetry, #performance, #nodejs, #motorsport, #api-design | Read Time: 11 min read

  • Binary Stream Parsing & Perceptual Color Extraction: Building @mikesven/gif-parser

    Most image processing in JavaScript depends on native C++ bindings like Sharp or libvips, which fail inside lightweight serverless isolates like Cloudflare Workers. Here is how I engineered @mikesven/gif-parser: a zero-dependency TypeScript engine that parses raw GIF87a/89a binary byte streams using DataView and implements CIELAB Delta-E color science for perceptually accurate palette extraction.

    Category: Systems & Binary | Tags: #typescript, #binary, #graphics, #color-science, #npm, #algorithms | Read Time: 11 min read

  • Deploying a High-Resolution Photography Platform: SST Ion, Cloudflare Pages, and Edge Caching

    Displaying fine-art wildlife and astrophotography on the web requires balancing 45-megapixel raw fidelity with sub-second page loads on mobile devices. Here is how I architected photography.mikesvendsen.net using SST Ion, Cloudflare Pages, static manifest extraction, and responsive AVIF/WebP srcset generation—delivering a zero-database gallery with global edge caching.

    Category: Edge Architecture | Tags: #cloudflare, #sst, #nextjs, #photography, #web-perf, #edge-architecture | Read Time: 11 min read

Systems & Open Source Projects

  • Reactiv E-Commerce Data & AI Platform

    Snowflake, dbt & AWS Bedrock AgentCore Platform for Shopify Brands

    Production analytics and multi-agent AI infrastructure supporting high-growth Shopify merchants. Orchestrates AWS (Bedrock AgentCore, Lambda, Step Functions, EventBridge, Glue, Redshift Serverless) and Snowflake with dbt modeling—establishing the unified, single-source-of-truth data layer that powers both merchant dashboards and autonomous AI Studio Agents.

    Tech Stack: Python, Snowflake, dbt, AWS Bedrock, AgentCore, Shopify Data, SQL

  • LiteMCP

    Cloudflare Worker Compatible Model Context Protocol (MCP) Framework

    Lightweight, edge-native implementation of Anthropic's Model Context Protocol (MCP) engineered specifically for serverless deployment on Cloudflare Workers. Features standards-compliant JSON-RPC 2.0 execution, real-time Server-Sent Events (SSE) streaming, CORS support, and zero-cold-start latency.

    Tech Stack: TypeScript, Cloudflare Workers, Model Context Protocol, JSON-RPC 2.0, SSE

  • GGUF-n-Go

    Automated Quantization Pipeline for Hugging Face Models to GGUF

    Your go-to automated tool for easily creating quantized versions of Hugging Face models in the GGUF format. Seamlessly handles modern architectures, tokenizer migrations (v3 tokenizers, consolidated safetensors), and exports optimized Q4_K_M, Q8_0, and FP16 weights ready for local llama.cpp and Ollama runtimes.

    Tech Stack: Python, Hugging Face, llama.cpp, GGUF, PyTorch

  • AetherCode & Fine-Tuning Labs

    Instruction-Tuned Code Models & Preference Alignment Datasets

    Open-source fine-tuned code LLMs and preference datasets published on Hugging Face. Includes Aether-Qwen2-0.5B-SFT (GGUF, GPTQ 4-bit, AWQ), Phi-nut-Butter-Codebagel, and the Reflective-MAGLLAMA dataset series (surpassing 10,000+ community downloads) trained with PEFT and SFTTrainer on dedicated Nvidia GPUs.

    Tech Stack: Python, PEFT, LoRA, Hugging Face TRL, SFTTrainer, Nvidia GPU

  • iracing-data-sdk

    Comprehensive TypeScript SDK for the iRacing Telemetry & Data API

    A production-grade, strongly typed TypeScript library for interacting with the iRacing Data API. Provides end-to-end typed schemas for all documented telemetry endpoints, automated chunked download handling, credential caching, and robust rate-limiting.

    Tech Stack: TypeScript, Node.js, REST API, Telemetry, NPM Package

  • ts-gif-parser (@mikesven/gif-parser)

    Binary Stream Parser & CIELAB Perceptual Color Table Extractor

    High-performance TypeScript utility for parsing raw binary information from .gif files. Extracts logical screen descriptors, binary headers, frame blocks, and derives exact RGB, Hex, and CIELAB perceptual color spaces directly from the global/local color tables.

    Tech Stack: TypeScript, Binary Streams, CIELAB Color Space, Nx Monorepo

  • spypoint_mcp

    Model Context Protocol Server for Cellular Trail Cameras

    Specialized Model Context Protocol server enabling LLM agents to monitor, query, and analyze photo capture streams from remote SpyPoint cellular trail cameras. Features 19 tools, 112 automated tests, read-only account security boundaries, and time-window clustering algorithms to group burst captures into discrete wildlife visits for predictive dawn scouting.

    Tech Stack: Python, MCP Protocol, IoT APIs, pytest (112 tests), Computer Vision

  • Michael Svendsen Photography

    Wildlife & Astrophotography Platform for Visual Storytelling

    Dedicated photography portfolio capturing the natural beauty and night skies of Ontario, Canada. Powered by SST Ion on Cloudflare Pages with edge caching, fluid gallery navigation, and Lightroom fine-art workflows.

    Tech Stack: Next.js, Cloudflare Pages, SST Ion, Tailwind CSS