Technical Publications & Systems Research

Everyone's racing to bolt AI agents onto products. But AI that recommends what to feature in a store is only as good as the numbers underneath it. Here is how we architected Snowflake and dbt so dashboards and AI agents consume the exact same verified numbers.

I wanted to ask my agent: 'Based on my trail-camera photos, when should I go photograph deer tomorrow?' So I built an MCP server for SPYPOINT with 19 tools, 112 automated tests, and burst clustering algorithms.

Engineering a lightweight, edge-native Model Context Protocol framework for Cloudflare Workers with real-time SSE streaming, zero cold-starts, and strict type safety.

How the Data & AI team at Reactiv leveraged Amazon Bedrock AgentCore and CopilotKit to build an autonomous Studio Agent that turns merchant goals into live, optimized mobile storefronts.

Why manually converting models to GGUF format is error-prone, and how GGUF-n-Go automates tokenizer synchronization, safetensor shard consolidation, and multi-precision quant matrix exports.

Architecting iracing-data-sdk: building end-to-end type safety, automated rate-limiting, and chunked download managers for real-time motorsport telemetry.

How I built @mikesven/gif-parser: a zero-dependency TypeScript library for byte-level parsing of GIF streams and extracting perceptual CIELAB color palettes.

How I architected photography.mikesvendsen.net: using SST Ion, Cloudflare Pages, and edge image optimization to showcase high-resolution fine-art photography with sub-second page loads.