Notes from the build
Everything I have learned the expensive way: Flutter in production, backend layouts that survive a rewrite, real-time systems, and what changes when you start leading a team instead of only shipping for it.
07 Oct 2026 · 16 min
Google ADK, drawn out: how agent systems actually work
Give an AI agent one real job and it becomes a team. Here is every piece of Google's Agent Development Kit that makes that work, each one run for real.
03 Oct 2026 · 15 min
LangGraph: what happens to your workflow when the server dies
We killed a real process in the middle of charging a card. With LangGraph's default setting the restart began from step one; with one flag it resumed at the charge. Either way, the charge ran twice. Here is the whole runtime: state, edges, reducers, checkpoints, interrupts, durability, parallel work, and what to ship.
30 Sep 2026 · 14 min
LangChain from zero: build your first AI agent, and make it safe to ship
An agent is a model, your functions, and a loop. Here are the eight building blocks of a LangChain 1.x agent, every one run for real, plus the three numbers that decide whether it is safe in production: 5,000 calls, 109k tokens, and one spoofed user.
22 Sep 2026 · 12 min
Where Jev actually goes in a system you already run
TypeSafe's System One model returns typed decisions with calibrated probabilities in under half a second. Six places to put it next to the LLM you already pay for — plus the confidence formula nobody explains, and where it breaks.
08 Sep 2026 · 9 min
From 114 audio files to a YouTube playlist: building a Quran waveform video pipeline
How I turned a folder of recitation MP3s into 114 calm, waveform-lit videos with Pillow, ffmpeg and a few hundred lines of bash — and what broke along the way.
12 Aug 2026 · 7 min
Why I still choose Flutter for production fintech
Four money apps in, the arguments against Flutter have mostly aged out — and the ones that remain are about platform channels, not the framework.
28 Jul 2026 · 6 min
The FastAPI service layout I use on every new project
Routers, services, repositories and one settings module — the folder structure that survived five products and three teams.
09 Jul 2026 · 5 min
Leading engineering without leaving the codebase
What changed when I became a CTO, what I got wrong about delegation, and the review habits that kept me useful to my team.
21 Jun 2026 · 7 min
Shipping an AI companion: what RAG got right, and where it lied
Building Riri taught me that retrieval quality is a product problem long before it's a vector-database problem.
02 Jun 2026 · 6 min
Multi-tenant from day one: lessons from an eCommerce SaaS
Shared schema, tenant column, row-level policies — and the three migrations that would have been impossible six months later.
14 May 2026 · 5 min
Agora, jitter buffers and the 3 a.m. call-quality bug
A week of debugging voice calls on rural 3G, and why the fix was in our reconnection logic rather than the SDK.
27 Apr 2026 · 5 min
Four Postgres indexes that fixed a 135,000-candidate search
Trigram, partial, composite and covering — how a recruitment platform's slowest screen became its fastest.
05 Apr 2026 · 6 min
Hiring your first three engineers
What I look for when there is no senior bench to hide behind, and why my take-home test is a paid, scoped, real ticket.
Build things that outlive the sprint.
The stack will keep changing — the framework you love today is tomorrow's legacy migration. What survives is the discipline: ship small, measure honestly, write the code you'd want to inherit, and mentor the person coming up behind you. Somewhere out there it's 2 a.m. and someone needs your software to just work. Build for them.