Off-Grid Survival Emotionally Intelligent Bot

Total time spent: 31h

28 June 2025

Spent: 1h

Ideas.. ideas.. ideas...

Idea: create animated face

29 June 2025

Spent: 2h

Today is DEADLINE.

Clarify the idea:

• use LLM and Raspberry Pi
• how about Gemma 3n from Google, there is 6B version which should fir 8GB Raspberry Pi 5
• we need text to speech, oh, well we can use Kokoro-82M
• we can animate the face and mouth with little servos
• let's also add 8x8 LEDs for eyes!

30 June 2025

Spent: 2h

• I asked Gemma 3n E2B

how to make shelter in the woods for 1 person

using text Gemma 3n E2B outputted:

Okay, here's how to build a basic shelter for one person in the woods:

1.  **Find a site:** Look for a protected spot – avoid low-lying areas prone to flooding. Consider wind direction.
2.  **Lean-to:** Find a sturdy branch or fallen tree. Lean branches against it at a 45-degree angle.
3.  **Cover:** Pile leaves, pine needles, ferns, or other debris over the branches for insulation and weather protection.
4.  **Insulate:** Add extra layers of insulation at the base.

**Stay dry & warm. Prioritize fire safety!**

Here is the audio using Kokoro-82M: audio

30 June 2025 (evening)

Spent: 3h

• Started learning how to do a CAD model!

• Squares at the top are eyes (8x8 LEDs)

• The mouth will be animated by

• The horizontal rod is a mouth :)

1 June 2025

Spent: 4h

• Successfully running Gemma 3n E2B with text, it works!
• Experiemnted with running Gemma 3n E2B with audio -- does not work
• Text-to-speech module (Kokoro) -- works!

2 June 2025

Spent: 2h

• Change of plans: instead of Gemma native audio, I'll use ASR+TTS and LLM for text only
• Experiments with Whisper-turbo

3 June 2025

Spent: 3h

• Trying to figure out how solar panel charging a batter would work
• Realized that BOM does not include battery (!)
• Got old Raspberry Pi 3, trying to setup 1B llama

4 June 2025

Spent: 2h

• Got Google Voice Kit (from my sister)
• Experiemnting with voice input/output

5 June 2025

Spent: 2h

• Got Kokoro TTS workin in INT8 locally using ONXX!!

See https://github.com/max1vol/emo-survaval-bot/blob/main/tts/tts.py And https://github.com/max1vol/emo-survaval-bot/blob/main/tts/audio.wav

6 June 2025

Spent: 6h

• LLM is working!
• Experimented with Qwen3-0.6B, it's a thinking model
• Experimented with Google's Gemma 3 1B Q4 quantization
• Aiming to run Gemma3-1B as thinking model might be not great for real-time chat
• Got ASR (automatic speech recognition) working using Whisper/tiny.en

7 June 2025

Spent: 4h

• Integrated ASR (speech recognition), LLM (text answering), and TTS (text to speech)
• The whole bot is working end-to-end!