Lightweight desktop app to index and query your meeting notes using a local LLM + vector store. It parses meeting notes (Markdown), chunks them, stores embeddings in a vector database (Postgres + pgvector), and provides a small Kivy UI for asking questions and running incremental indexing.

• Parse meeting notes from a directory (configurable).
• Chunk text, extract metadata (title, date, attendees, action items).
• Store embeddings in Postgres via pgvector.
• Use a local or remote LLM (Ollama or other) through LangChain for Q&A chains.
• Simple Kivy-based UI for ad-hoc queries and indexing operations.

• Code: this repo
• Config: settings.json
• Env helpers: env_setup.sh, README_ENV.md
• Pre-download helper: download_hf_model.py
• Sample meeting note: meeting_2025-09-17_Sample_Meeting_Title.md

• Python 3.11+
• See requirements.txt for the Python dependencies (LangChain, Kivy, sentence-transformers, SQLAlchemy, psycopg2-binary, etc.).

1. Create and activate a virtualenv:

python3.11 -m venv .venv
source .venv/bin/activate

2. Install dependencies:

pip install --upgrade pip
pip install -r requirements.txt

3. Configure environment variables. You can use the provided env_setup.sh to load values from a .env file:

cp env_setup.sh .env          # edit .env and set vars like PG_CONN, OLLAMA_HOST, HF_MODEL_DIR
source env_setup.sh

4. (Optional) Pre-download the Hugging Face embeddings model to avoid repeated downloads:

python download_hf_model.py --model sentence-transformers/all-MiniLM-L6-v2 --out ~/.cache/hf/all-MiniLM-L6-v2
export HF_MODEL_DIR=$HOME/.cache/hf/all-MiniLM-L6-v2

5. Run the app:

python main.py

• settings.json contains the default path for your meeting notes directory. Current value:

meeting_notes_dir: "/mnt/c/Users/vswam/Obsidian/Work/Diary/"

Override configuration using environment variables (preferred):

• PG_CONN — Postgres connection string (required for pgvector storage)
• OLLAMA_HOST — URL for Ollama if using a local LLM
• MODEL_NAME — model name for Ollama/LLM
• HF_MODEL_DIR — local path to a pre-downloaded HF embeddings model

Check README_ENV.md for more detailed steps on environment setup and troubleshooting.

• The codebase uses lazy initialization for heavy runtime resources (embeddings, LLM, vector store) to avoid import-time failures.
• note_parser.py contains the parsing logic for meeting notes — it extracts title, date, attendees, bullets, and action items.
• indexer.py performs incremental indexing and writes to the vector store.
• rag_pipeline.py wires embeddings, vector store, and the chain used by the UI.

• If imports fail at runtime, make sure you installed the dependencies in an activated virtual environment.
• If the app can't reach the vector store, verify PG_CONN and that the Postgres server has the pgvector extension installed.
• To check DB connectivity, use the included check_pg_conn.py script.
• If embeddings appear to re-download each run, set HF_MODEL_DIR to a local copy of the model (see download_hf_model.py).

• See LICENSE.

• Venkat Swaminathan (c) 2025