Why I Rebuilt My Lab Pipeline Twice Before It Was Right
raw posts leaked real client names; a dual-model denylist was too blunt; I landed on a deterministic scrubmap + Sonnet generalize pass + a per-post g
A lab notebook: what I'm learning while building with AI, written as I go.
raw posts leaked real client names; a dual-model denylist was too blunt; I landed on a deterministic scrubmap + Sonnet generalize pass + a per-post g
Don't strip the quoted chain. Don't dump it. Fold it. The message thread is your context — but only the reader head needs to be readable on arriva
I built a feedback loop into my AI dashboard before I built the inspection surface. When I finally could see what it had learned, I immediately wipe
If the people in the meeting make decisions per PO, your dashboard row has to be a PO. Grouping by product looks tidier. It's lying.
Placeholder data tests your assumptions. Real training data tests your code. They are not the same test.
My automated CEO briefing kept dying at exactly 8 AM. The logic was fine — it was the clock. Reproduce off-peak first. If it works, you have a conte
Your DNS resolver may be lying, and that first-year .co price definitely is. Here's the two-check method I now run on every domain buy.
Apollo, my Claude-backed dev agent, was booting with full identity and rules — but zero awareness of what we'd shipped last session. A single rollin
If you're using AI agents to build, your value isn't in the keystrokes. Price the orchestration, not the labor.
There's no single 'make PDF' command. Match the pipeline to the content shape — image-composed pages get Pillow + img2pdf, designed HTML docs get he
A health check that says ok tells you the process is alive — not that it's running the code you think it is. Make it report identity.
Two organizing axes = everything shows up twice. Pick ONE. Then delete.
The Setup
Before you touch a committed sprite (or any binary asset), run git log -- <path>. The generator script sitting in your repo is a hypothesis, not
We gated Phase 3 of an Apollo memory build on a real benchmark before writing a single line of reranker code. The gate came back no-go. That was the bes
FreeSWITCH is the open-source engine. SignalWire is the managed cloud built by the same people who wrote it — with a Twilio-compatible API bolted on. On
Pillow's multipage PDF save crashes if your Python build has no JPEG codec. img2pdf + getfixeddpilayoutfun is the fix. And every build that hits
A well-built leave-behind isn't a brochure. It's an instrument. Every inclusion is a deliberate choice — and every omission is too.
What I Did on Purpose
Twilio's console has at least three places where the UI says 'success,' nothing actually happened, and there's no error. Verify state. Don't trust absen
When a client finds a real competitor, resist the urge to strawman it. Do an honest teardown. That's where the actual wedge lives.
the setup
The Setup
A script named gensprites.py in the repo looked like the source of truth for the assets. It was the thing that got replaced. We ran it anyway,
If your pitch leads with what you do and ends with the number, you've got it backwards. Start with how you show up. Defer the price. Niche until it stin
My first leveling formula hit Level 12 in 11 weeks. That's not a progression system — it's a countdown clock.
Two layers, one store. Deterministic for instant. Batch LLM for generalization. Never re-run the pipeline on every thumb.
30908 privacy policy can't be verified is almost never about your privacy page wording. It's about whether a human vetter can physically reach your op
Tailwind v4 changed translate utilities to emit the CSS translate: property instead of transform:. If your animation keyframe uses transform, the
My RAG system silently ignores PDFs. Every PDF I'd ever dropped in was invisible — no error, just gone. The fix: extract to .md, clean the junk, reind