Backend and request engineering, mostly Go. High-concurrency automation, low-latency monitoring, and the full-stack tooling that operates it.

• flatread · read FlatBuffers buffers when you don't have the schema. Field access by vtable slot instead of by generated name, for plain and size-prefixed buffers alike. No dependencies, fuzz-tested so it never panics on malformed input.
• proxypool · rotating proxy pool with cooldowns. Small: no health checks, because only your own traffic can tell you whether an address works.
• hardiff · compare two HTTP Archive captures. Matches requests by endpoint rather than by full URL, so a cache buster does not turn every call into one request removed and one added, and compares JSON bodies field by field instead of as a blob.
• flatschema · infer a FlatBuffers schema from sample buffers. Merges what each sample populated and writes the .fbs they imply, keeping the slot numbering right where a field nothing populated would otherwise shift every field after it.
• identlint · check that a request's headers agree with the browser they claim to be: the version in User-Agent against sec-ch-ua, the arbitrary brand Chromium puts in that list and where, the platform and mobile hints, the encodings, a TLS profile name. The brand rules are Chromium's own and were checked against Chrome, Edge and Brave on the wire. Reads a header block or a whole HAR.

• bogdanfinn/tls-client#266 · a cached transport was reused after the server changed the protocol it negotiated, so an HTTP/1 transport read an HTTP/2 SETTINGS frame as a status line. Merged, in v1.16.0.
• bogdanfinn/tls-client#270 · the transport cache was written from a reconnecting dial holding a different mutex than the one guarding it, so a read and a write could overlap. Found with the race detector. Merged.
• bogdanfinn/fhttp#27 · a deflate body was sniffed and buffered while the response was still being built, so over HTTP/2 the request never returned. gzip and brotli on the same host were fine. Merged.

Go · TypeScript · Node · React / Next.js · PostgreSQL · MongoDB · Redis · Docker · REST & GraphQL

Request-based automation in Go: site modules, CLI tooling, and systems that have to stay fast under heavy concurrency. Real-time Discord tooling, scraping and database sync. Monitoring with sub-second reaction times. Dashboards and API backends to drive all of it.

field-notes · what I learned from being confidently wrong 115 times while reverse engineering a bot-protected site. 110 of them were my own bug.

Available for project work and monthly retainers. Reach me at info@mh-automation.nl.