// transmission log
Thoughts, write-ups, and things worth documenting.
Built and tuned a self-hosted LLM inference server on a single consumer GPU — nearly tripling its throughput with smarter model placement — then made it privately shareable with a team over an encrypted mesh network with key-based auth.
read more →The next leap in AI-augmented work isn't a better model. It's a better system around the model. Here's the architecture we built — and why Claude Haiku inside it outperforms raw Claude Opus on the same tasks.
read more →The ServiceNow AI chat bot never asks Claude to figure out what data exists. It fetches first, structures the result, then hands it to the model. That boundary is the architecture — and it's why the bot works reliably.
read more →A security scan flagged missing HSTS, CSP, X-Frame-Options, and three other headers on a static site hosted on Namecheap Apache. Here's the exact .htaccess fix and why each header matters.
read more →Took the AI chat widget from a working prototype to an enterprise-ready app — adding four role-based tiers, L1 troubleshooting flows, write actions, audit logging, rate limiting, and CIS-compliant credential storage.
read more →sys_ui_script with global=true sounds like it should work everywhere. It doesn't. Here's the exact diagnostic, the correct injection pattern for Service Portal, and three other bugs we hit building a Claude-powered chat widget.
read more →Designed and deployed a Claude-powered self-service chat bot inside ServiceNow — giving employees instant answers about their tickets, requests, and catalog items without opening a single form.
read more →The precise ServiceNow object stack for wiring a live external API into a catalog form — REST Message, two Script Includes, a client script, and the GlideAjax bridge that ties it together.
read more →Designed and built a ServiceNow catalog item that queries a banking API in real time — auto-populating customer name and account type as the user types, with automatic routing to the fulfillment team.
read more →A technical and executive breakdown of what becomes possible when an AI system has direct, live access to your ServiceNow instance — and why it matters for teams at every level.
read more →How we established a stable MCP connection to ServiceNow, benchmarked an entire instance, and built a live intelligence layer on top of it.
read more →How I built PAI — a 7-phase AI agent system with 49 skills, voice interface, and persistent memory on top of Claude Code.
read more →Two Python CLI tools — pdf2md and paper2md — for converting books, reports, and research papers into clean LLM-ready Markdown.
read more →How we debugged three silent failure layers to establish a live MCP connection — then used it to configure actual ServiceNow records without a browser.
read more →How to integrate a banking API with ServiceNow catalog items using REST messages, GlideAjax, and multi-type fallback patterns.
read more →Why ServiceNow catalog items silently fail to create RITMs and how to trace onSubmit client scripts.
read more →Why ServiceNow workspaces show blank screens and how variant audience matching and sort order affect visibility.
read more →How we built an automated workflow to publish project posts directly from our AI workflow.
read more →