Create personal notes with voice transcription using local LLaMA and Telegram

PersonalNotesAssistant – Organize and Understand Your Thoughts with Local AI

PersonalNotesAssistant is an offline-capable, AI-powered agent that helps you store, summarize, retrieve, and reflect on your personal notes and voice memos — all processed locally and sent via Telegram. Built to run efficiently on a Raspberry Pi 5, this agent supports a variety of note-taking styles and acts as your private memory extension.

🧠 What It Can Do

Accept voice or text notes via Telegram

Transcribe audio messages into clean, structured text (using Whisper)

Automatically summarize or categorize notes with a local LLM

Answer questions based on your past notes

Retrieve relevant entries by topic, date, or keyword

Help you journal or reflect by asking follow-up questions

Work completely offline — no cloud or external APIs

🔧 How It Works

Capture Notes via Telegram You send a voice message or text to your Telegram bot. The assistant supports both quick thoughts and long-form content.

Transcription with Whisper (Local)

If the input is a voice message, it is transcribed into text using Whisper running locally on your Raspberry Pi.

AI Summarization & Tagging

The transcribed or typed note is sent to LLaMA 3.2 via Ollama, which summarizes it, suggests tags, and stores it with metadata (e.g., timestamp, mood, theme).

Storage & Retrieval

Notes are stored in a local database (e.g., SQLite or JSON). You can later query the assistant with prompts like:

“What did I say about stress last week?”

“Summarize my ideas from this month.”

“Show notes tagged with 'travel'.”

Follow-Up & Reflection The agent can optionally engage with reflective prompts to help you deepen your thoughts or gain insight from what you’ve recorded.

💡 Use Cases

Track personal growth, habits, or therapy progress

Create voice memos while walking or commuting

Maintain a structured journal without typing

Use as a second brain to help you remember and revisit important thoughts

🔐 Privacy by Default

Everything runs locally:

No notes are uploaded to cloud platforms

No audio is sent to third-party transcription services

No LLM processing happens outside your device

Ideal for privacy-minded users, psychologists, researchers, or digital minimalists who want AI assistance without surveillance.

⚙️ Technical Stack

Raspberry Pi 5: Low-power edge device

Whisper (local): For voice-to-text conversion

Ollama + LLaMA 3.2: For summarization, classification, and retrieval

Telegram Bot API: For input/output

Custom Database (e.g., JSON/SQLite): For storing and querying notes

🧪 Real-Life Use

This agent is actively used daily by the developer to log ideas, emotions, and plans. It has proven effective for lightweight journaling and context-aware memory assistance, even when offline.