Blog
Build in Public
October 3, 20259 min

Why I created a voice task manager with AI

TAMSIV was born from a common frustration: grocery lists stuck to the fridge, a diving club managing everything on WhatsApp, and that constant feeling that information gets lost between post-its and chat groups. One day, I thought: why can't we just talk to organize our lives? Six months and 700+ commits later, TAMSIV is a complete Android app with conversational AI, real-time voice pipeline, and collaborative management. Here's how it all started.

Key takeaways:
- The best product ideas come from repeated personal frustrations — not abstract market research.
- React Native with the New Architecture (Fabric) allows a solo web developer to deliver a high-performance mobile app without pure native code.
- Supabase offers everything a solo project needs (auth, DB, storage, Realtime) in a single service.
- The first functional prototype (voice to created task) is the best idea validator — better than any business plan.
Refrigerator door covered with handwritten grocery lists and colorful post-it notes
The reality of family task management: scraps of paper on the fridge that get lost every week.

Where did the idea for a voice-controlled task manager come from?

It all started with a common frustration. My wife and I manage a family of four. Grocery lists? On a piece of paper stuck to the fridge. Household chores? In our heads. And when we forget something — which happens often — it's that little daily tension. "Didn't you say you'd take care of it?" Yes, but I forgot. Like every time.

In parallel, I'm a member of a diving club. And there, it's organized chaos: everything goes through WhatsApp. Outings, registrations, equipment, reports. Dozens of messages a day in a group where information gets drowned out. Are you looking for the date of the next outing? Good luck finding it between the memes and fish photos.

Two different contexts, the same problem: information gets lost because the tools we use aren't designed for it. Paper doesn't remind you. WhatsApp doesn't organize. And existing task management apps? Too complicated, too many fields to fill, too much friction.

Why is voice the answer to mental overload?

The insight that started it all: we think and communicate through speech, not forms. When you tell your spouse "don't forget to buy milk tomorrow," it's a clear instruction. But when you open a task app, you have to:

  1. Open the app
  2. Tap "+"
  3. Type the title
  4. Select a date
  5. Choose a project/tag
  6. Save

Six steps for an action that takes 3 seconds to say. This is exactly the problem described by the concept of capture friction in David Allen's GTD (Getting Things Done) methodology: the easier it is to capture an idea, the more the system is used.

I thought: why can't we just say what we want to do? "Add milk to the grocery list." "Remind me to prepare the bottles for Saturday." Without opening an app, without typing, without searching for the right field in a form.

That's how TAMSIV was born. A task and memo manager driven by voice, with conversational AI that understands what you want and acts accordingly.

What technical choices for a solo developer launching a mobile project?

Developer setting up a new project on a laptop with the first lines of code
The first commit: October 2025. A monorepo, three folders, and a mountain of decisions to make.

First commit in October 2025. Every technical choice at this stage is critical — you're alone, you have no margin for error, and going back costs a lot.

Why React Native and not pure native?

I'm a web developer by training. Pure native (Kotlin/Swift) would have slowed me down by several months for an equivalent result. React Native with the New Architecture (Fabric) delivers on performance — the jank and bridge issues that plagued earlier versions are resolved.

The choice of a monorepo with three folders was obvious:

  • frontend/ — React Native TypeScript for the mobile app
  • backend/ — Node.js Express with WebSocket for the AI server
  • website/ — Next.js for the landing page and web dashboard

One repo, one git log, shared TypeScript types between frontend and backend. Consistency is maximized.

Why Supabase rather than Firebase or a custom backend?

Supabase offers everything a solo project needs in a single service: managed PostgreSQL, integrated authentication (email, magic links, and later QR code), file storage, and Realtime for real-time synchronization.

Firebase would have been the other obvious option, but the NoSQL model (Firestore) is not suitable for relational data like tasks, memos, groups, and events. PostgreSQL is made for that — and Supabase makes it accessible without managing a server.

The database structuring would come later, but choosing Supabase from the start allowed for fast iteration without worrying about infrastructure.

How does the voice pipeline at the heart of TAMSIV work?

Person speaking into their smartphone while walking outdoors
The ideal use case: saying what you want to do, hands-free, without opening a menu or filling out a form.

The core of the project is the voice pipeline. From the first prototype, I wired up the complete chain:

  1. PCM 16kHz Audio — captured from the phone's microphone
  2. Real-time WebSocket — sending the audio stream to the backend
  3. STT (Speech-to-Text) — transcription by the device's native engine (with Deepgram fallback)
  4. LLM (Large Language Model) — text analysis and action decision via OpenRouter
  5. Function calling — the LLM calls a function (create_task, create_memo, etc.)
  6. TTS (Text-to-Speech) — voice response via OpenAI TTS

The user speaks, the backend transcribes, sends the text to an LLM which decides what to do, and returns a voice response. All in a few seconds.

The first tests were magical. Saying "create a task buy bread for tomorrow" and seeing the task appear with the correct date... that's the kind of moment that confirms you're on the right track. No need for a 50-page business plan — when the prototype works and you think "I want to use this every day," the idea is validated.

What are the challenges for a solo developer on such an ambitious project?

Being a solo developer means total freedom on technical choices. No meetings, no compromises, no "we'll discuss it on Monday." You see a problem, you fix it. You have an idea, you code it.

But it's also the loneliness when facing bugs at 11 PM. No one to review your code, no one to challenge your choices, no one to tell you "you're on the right track" when you doubt yourself. It's a feeling many solo technical founders know.

Building in public — sharing the journey through this blog and social media — has become my way of compensating for this loneliness. Each article is a checkpoint, proof that the project is moving forward, a way to receive feedback even without a team.

How did we go from the first commit to 700+ commits in 6 months?

The journey since that first commit followed a trajectory I couldn't have planned:

The detailed article on the 650+ commits traces each step in depth.

What lessons can be learned for launching a solo project?

  1. Start from your own frustration. The best products solve a problem the creator experiences daily. You will be your first user and your best tester.
  2. Choose tools that multiply your productivity. React Native, Supabase, OpenRouter — every choice should save you weeks, not hours.
  3. Prototype fast, validate fast. The first functional voice pipeline took me 3 days. This prototype validated the idea better than any study.
  4. Document the journey. Building in public isn't marketing — it's a logbook that forces you to step back and creates a community around the project.

Frequently Asked Questions

Do you need to know native coding to create a mobile app with AI?

No. React Native with TypeScript allows a web developer to create a high-performance mobile app. For AI, services like OpenRouter and OpenAI provide APIs that don't require machine learning skills. The most important thing is to understand client-server architecture and WebSockets.

How much does it cost to develop an app like TAMSIV as a solo developer?

In direct costs: Supabase (free plan initially, then ~25 EUR/month), OpenRouter (~0.001 EUR/LLM message), OpenAI TTS (~0.015 EUR/1000 characters), Runware for images (~0.003 EUR/image). The main cost is time: 6 months part-time, or about 500-600 hours of work.

Why a monorepo rather than separate repos?

For a solo developer, consistency is paramount. A single repo means a single git history, shared TypeScript types, and no version synchronization between packages. A monorepo becomes more complex to manage with 3-4 developers, but for a solo project, it's ideal.

Does the voice pipeline work offline?

Partially. The device's native STT works without internet (local speech recognition). But the LLM and TTS require a connection to the backend. An offline mode with queuing is on the roadmap — voice commands would be stored locally and processed upon reconnection.

Is WhatsApp not enough for group task management?

No, and that's exactly the problem that inspired TAMSIV. WhatsApp is a communication tool, not an organization tool. Information gets lost in the message stream, nothing is structured, there are no reminders, no due dates, no assignments. It's organized chaos — and TAMSIV's collaborative groups precisely solve this problem.

All articlesNext article