Lookout: The Offline AI Field Companion That Tells You to Put Your Phone Away

┌─────────────┐
│ Ask once    │
└──────┬──────┘
       ▼
┌─────────────────────────┐
│ One fact + one mission  │
└──────┬──────────────────┘
       ▼
┌─────────────────────────┐
│ Put the phone away      │
│ and observe the world   │
└──────┬──────────────────┘
       ▼
┌─────────────────────────┐
│ Return and save notes   │
│ as a field journal      │
└─────────────────────────┘



The target is **less than 60 seconds of total screen time per session.**

- Source code: Lookout on GitHub — `https://github.com/aun009/hacktober-week-1`

A good demo flow is:
1. Open Lookout in **Airplane Mode**.
2. Ask a typed or offline speech question.
3. Show the one-fact response and observation mission.
4. Return to the app, add a journal note, and export the PDF.

## Code

The project is open source under the **Apache 2.0 License**.

Key parts of the project:
- Android app and Compose UI
- Offline species-pack retrieval
- Mission system
- Safety rules
- Field journal and PDF export
- Architecture documentation

You can embed the repo directly with:

{% github `[your-github-username]/[repo-name]` %}

## How I Built It

Lookout is built with **Kotlin and Jetpack Compose** for Android. Its core design avoids a cloud-dependent loop: local field notes, missions, journal entries, and PDF export work without Lookout sending data to a server.

The repository also contains interfaces and architecture for a fully local open-source AI stack:

| Capability | Open technology planned / integrated architecture |
|---|---|
| Local language model | Qwen via llama.cpp |
| Speech-to-text | whisper.cpp |
| Text-to-speech | Piper |
| Voice activity detection | Silero VAD |
| Noise reduction | RNNoise |
| Image understanding | MobileCLIP |
| Local data | Room / SQLite |
| UI | Kotlin + Jetpack Compose |

I deliberately designed the app to **fail safely**. If a model-backed answer is unavailable, Lookout says so rather than inventing an identification. The current build does not bundle pretrained model weights or enable live camera and ambient bird-call identification yet; those are clearly documented as next steps in the repository.

Safety is especially important outdoors. Lookout's rule layer flags potentially dangerous situations involving toxic plants, mushrooms, or wildlife and avoids presenting an AI response as permission to eat, touch, or approach something.

## Why Does Open Innovation Matter?

Open innovation is what makes Lookout's purpose possible.

A closed, cloud-only AI API would make an outdoor companion depend on signal, credentials, recurring cost, and a server connection. That is exactly the wrong trade-off for a trail, park, forest, or rural area.

Open-source AI makes it possible to build toward:

- **Offline use:** Local models can run where internet access is unreliable or unavailable.
- **Privacy by architecture:** Voice and observation data can stay on the device instead of being uploaded to a third-party server.
- **Model choice:** The architecture is designed so models can be swapped rather than locking the app to one provider.
- **Transparency:** The safety rules, species-pack workflow, and system behavior are visible in the codebase.
- **Lower barriers to experimentation:** Builders can adapt Lookout for a local region, language, or species pack without needing permission from a closed API provider.

For this project, open source is not just a licensing choice. It is the reason the product can aim to be **private, resilient, adaptable, and useful outdoors.**

## Prize Categories

I am entering the overall **Hacktoberfest Open-Source AI Challenge: Week 1 — Touch Grass** category.

I am not claiming a partner-technology category because this version does not use a listed partner technology.

## Team

Built by [@aun009](https://dev.to/aun009).

---

Thanks for reading — and if you try Lookout, I hope it gives you a reason to look up from your phone.