Orma - Accountability through Calls

A daily accountability call that remembers what I keep not doing.

Comment

Inspiration

I am a forgetful person. I have tried task apps, reminders, calendars, and notes, yet I still forget what matters.

The problem is not a shortage of reminders. It is how easily I dismiss one without telling anyone.

I wanted a gentler form of accountability. One that calls back, remembers the pattern, and lets me speak naturally.

What it does

Orma is a personal accountability call that turns a small spoken check-in into a memory I can return to.

I choose a time, then Orma calls. It asks what I am avoiding, what changed, or whether an unfinished task deserves another day.

It remembers repeated mentions and unresolved items. I can also say that something is done, paused, or no longer relevant.

The name means memory in Malayalam. Its identity starts from the Malayalam long O, then becomes the O in Orma.

Technology stack

The web app uses SvelteKit on Cloudflare Workers. Supabase provides Postgres, Auth, Storage, pg_cron, and Edge Functions.

CALL-E runs the live phone conversation. Gemini through Vertex AI transcribes and phrases it, while SQL computes every count and age.

The call pipeline claims work with FOR UPDATE SKIP LOCKED and uses an idempotency key. Those two safeguards prevent a person receiving the same call twice.

Orma also includes Telegram sign-in and notifications, Resend email, and an MCP surface for reviewing the accountability history.

The frontend remains deliberately small. The important work happens in the call lifecycle, where each run is claimed, briefed, dialled, transcribed, and recorded safely.

Challenges we ran into

The hard part was making a phone call feel caring without becoming intrusive. A reminder can be ignored, while a voice carries more pressure.

We designed each call around consent, a selected time, and a short exit path. Saying that an item is done or irrelevant removes it from future check-ins.

The other challenge was trust. The model never invents counts or ages. SQL calculates those values before the call receives a flattened briefing.

What we learned

The useful unit is not a task. It is a repeated hesitation that deserves a little reflection.

Voice makes the interaction feel human, but reliability matters more than charm. An incorrect count on a call would damage the product immediately.

What is next

We want to make the review history clearer, improve pattern reports, and let people choose the kind of accountability that helps them most.

We will keep the product small and consent-led. Orma should feel like a trusted check-in, not another system demanding attention.

Demo disclosure

The demo records a real Orma call. The on-screen history is seeded dry-run data and will be labelled as such.

Built With

Share this project:

Updates