FlowMem

Inspiration

In today's fast-paced world, we juggle countless responsibilities—finances, daily tasks, mental health, and personal reflections. Using five different apps to track all of this is tedious and causes friction. We realized that voice is the fastest medium to capture a thought, and with recent advancements in AI, we can seamlessly turn unstructured voice dumps into perfectly organized, actionable data.

What it does

FlowMem is a voice-first, intelligent personal companion and dashboard. By simply tapping a microphone and speaking your mind, FlowMem automatically parses your words and categorizes them into:

Tasks: Automatically detects deadlines, sets priority levels, and schedules aggressive background alarms so you never miss them.
Expenses: Extracts monetary values and categories to keep a running total of your daily spending.
Health & Mood Logs: Tracks your well-being, automatically mapping your feelings to intuitive emojis (😊 for good, 😔 for bad, 😴 for tired).
Journal: Transcribes and summarizes longer thoughts into a searchable personal diary.

All of this is presented in a stunning, minimalist dark-mode dashboard inspired by Vercel's design language.

How we built it

Frontend: We built the mobile application using Flutter & Dart, focusing heavily on a premium, responsive UI experience.
Backend: We used a Python backend hosted on Render to process the audio, interact with the LLM for categorization, and trigger push notifications.
Database & Messaging: Firebase Firestore handles real-time data syncing, while Firebase Cloud Messaging (FCM) powers our remote data payloads.
Native Integration: We utilized flutter_local_notifications with advanced Android core library desugaring to handle complex background alarms.

Challenges we ran into

Our biggest hurdle was the Android notification system. We didn't want standard, quiet "push notifications" for tasks—we wanted a real, aggressive alarm clock that users couldn't ignore.

Modern Android devices aggressively suppress background activity and silence notifications when phones are on vibrate or "Do Not Disturb." To bypass this, we had to dig deep into native Android AudioAttributes, create custom notification channels mapped directly to the user's Alarm Audio Stream, and implement a FLAG_INSISTENT payload that forces the phone to continuously ring and vibrate until the user manually acknowledges it.

Accomplishments that we're proud of

The Voice-to-Action Pipeline: It feels like magic. You just speak naturally, and the dashboard instantly updates with categorized, actionable items.
The Aggressive Alarm System: Engineering a notification system that reliably wakes up the device and bypasses silent modes from a background isolate was a huge native-code victory.
The Aesthetic: We are incredibly proud of the UI. It looks and feels like a premium, production-ready product.

What we learned

We learned an immense amount about bridging the gap between cross-platform frameworks (Flutter) and deeply native APIs (Android Notification Channels, Java 8 Desugaring, Background Isolates). We also learned how to effectively prompt and parse LLM outputs to reliably extract structured JSON data (tasks, expenses, mood) from completely unstructured audio transcripts.

What's next for FlowMem

iOS Support: Bringing our native alarm implementations over to APNs and iOS.
Proactive AI Advice: Using the historical data (expenses + mood + journal) to offer personalized insights (e.g., "You tend to spend more on dining when your health log says you are stressed").
Calendar Integrations: Automatically syncing parsed tasks to Google Calendar.