17 yrs shipping production software & architecture.
Building OpenPrompt OS (open-source): self-healing AI prompt studio.
Sharing systems lessons & code.
After 17 years of designing production systems and writing code for other companies, and lurking on here since 2009, I’m finally breaking my silence.
I’m using my nights and weekends to build OpenPrompt OS part-time, and I’m building it 100% open-source in public.
@manoj_ahi Which model are u using?
I bought a wired headset from the same brand from flipkart, which turned out to be fake product while i tried to raise some product complaint on their offical website
@thelifeofrishi Knowing what to build saves more time than shipping fast. We are building the prompt generator which will solve this problem. We are working on it and launching soon.
Spent the morning reading about that breakthrough pancreatic cancer pill on HN.
Wild what happens when researchers spend decades debugging biological state machines instead of arguing about frontend frameworks.
Before: Forcing half-baked Passkey flows into the primary login dialog and dealing with endless onboarding edge cases.
Now: Clean Turnstile-protected Email OTP + OAuth. Passkey schema stays in the database for later.
Spent the evening trimming token overhead in our prompt compiler because grid power availability for data centers is actually becoming a local dev bottleneck.
If we keep bloating context windows with unnecessary boilerplate, nobody's local setup will survive 2028 energy caps.
Shipped raw copy-pasteable prompt output in OpenPrompt OS today.
Killed all the template boilerplate wrapping our generated text in apps/api and guided-parameters.tsx so you can drop outputs straight into your LLM call without manual cleanup.
Background agent work rarely needs an instant answer, so a deferred tier is an easy win. In a Java service I would split model calls into interactive and batch paths, send nightly summaries and eval runs to the cheap tier, and keep timeouts and retries separate for each.