what i've shipped.

manage and monitor llm service accounts — cli, swift library, and macos menu bar app.

telegram bot — send a link, get structured metadata and a searchable archive.

incident summarizer shipping real customer text through azure openai — with a tiered pii guard at the chokepoint and a self-serve secure-llm-use pattern that replaced per-team security review.

gcp data pipeline — terragrunt-managed ingest and transform with dataproc, bigquery, pub/sub.

multi-cloud terraform — same patterns across aws, azure, gcp.

reference architecture for a cloud-native multi-tenant platform on eks — 99.99% sla target.

terraform — aws vpc + subnets in us-west-1 and vmware vcenter 7.0 vms from one codebase.

list, save, and download blobs from azure storage — python and bash flavors.

powershell scripts for veeam backup 365 and microsoft 365 administration.

log processor — extracts receive and saving rates from support logs, plots with matplotlib.

knowledge base article inventory — crawls xwiki, tracks changes via github actions.

openai-shaped proxy in front of claude — translation, microbatching, and a cost & latency observatory.

telegram bot + CLI for AI image generation — provider-pluggable across azure openai and gemini behind one interface.

interactive 3D car explorer — click a part, get a live LLM-generated explanation of what it is and why it matters.

tiered redaction gateway for log bundles on their way to an llm provider — presidio patterns, local ner over the residue, and a proxy that refuses to forward what it could not clean.

multi-agent runtime with a real separation between advisory and acting traffic — plan → human approval → tools → publish, holds as durable graph interrupts, not ui checkboxes.

fusion layer over two security vendors that do not talk to each other — ranks findings by blast radius in your estate, and no suppression happens without passing a deterministic check the model cannot argue with.