I Built ChronoDump: An AI That Turns Thought Dumps into Reminders

Ghananil Shirpurkar — Sat, 03 Oct 2026 17:02:49 +0000

This is a submission for the Hacktoberfest Weekend Challenge: Build for a Friend

What I Built

I built ChronoDump, a local-first AI Telegram bot for people who have a lot of thoughts, tasks, and deadlines but don't want to stop and organize them manually.

The idea came from a simple problem: sometimes you don't need another productivity app — you just need somewhere to dump your thoughts.

You can send ChronoDump a messy voice note or text such as:

“I need to finish chapter 4 tonight by 10, the report is due on 3rd January, remind me to drink water in two minutes, and I need to pick up my laundry sometime this weekend.”

ChronoDump transcribes and understands the dump, turns it into structured notes and action items, extracts deadlines, resolves time expressions, and automatically arms reminders.

When a time expression is ambiguous, it doesn't guess. Instead, it asks for clarification directly in Telegram.

The goal was to build something that feels like a small personal assistant rather than another task-management interface.

Demo

The working project is demonstrated through screenshots of the Telegram bot.

The screenshots show:

Voice note → structured notes
Extracted action items
Automatically armed reminders
Ambiguous-time clarification
Reminder notifications
+30m snooze
Marking reminders as done
Undo after completing a reminder
Direct text input

Code

GitHub: https://github.com/GhananilShirpurkar/ChronoDump

How I Built It

ChronoDump is built around local/open-weight AI rather than a closed cloud API.

The pipeline is:

Telegram → Whisper → Qwen3 → Pydantic → Temporal Engine → APScheduler → SQLite → Telegram

Stack

Python
aiogram 3 — Telegram bot framework
faster-whisper — local voice transcription
Qwen3 4B + Ollama — local language understanding
Pydantic v2 — structured model output
Python datetime + zoneinfo — deterministic time handling
APScheduler — reminder scheduling
SQLite — local persistence
SQLAlchemy — database access

The core design principle is:

LLM interprets. Python validates and reasons about time. APScheduler executes. SQLite remembers.

The LLM understands the user's intent, while deterministic Python code handles actual timestamps and scheduling. This prevents the model from directly deciding when a reminder should fire.

ChronoDump also includes a clarification flow for vague expressions such as “this weekend” or “later today” instead of silently making assumptions.

Why Does Open Innovation Matter?

ChronoDump is designed around the idea that useful AI doesn't always need to live behind a proprietary API.

Using open-weight models and local inference makes it possible to:

Keep personal voice notes and task data local
Avoid sending private dumps to a third-party AI provider
Run without relying on paid inference APIs
Swap models as better open models become available
Inspect and modify the AI pipeline
Self-host the entire system

For a tool that processes personal thoughts, reminders, deadlines, and voice notes, privacy and control are part of the product itself.

Open innovation made it possible to build this assistant using components that can be inspected, replaced, and self-hosted rather than locking the project to a single proprietary AI provider.

Prize Categories

Build for a Friend
Open-source AI / Open Innovation

DEV Community: Ghananil Shirpurkar