Loom link: https://www.loom.com/share/6ffc496cf8e5476abe2ee84b3193d168

Harper Insurance Memory System Dashboard

Streamlit dashboard for exploring brokerage accounts, events, semantic search, and follow-up planning backed by Postgres + pgvector.

Features

• 📊 Dashboard KPIs and activity charts
• 🔍 Natural-language query engine over events
• 👥 Accounts browser and event timeline
• 📅 Follow-up agent plan generation
• 🗄️ Database schema browser

Requirements

• Python 3.10+
• PostgreSQL with pgvector extension
• API keys for Anthropic and Voyage (for LLM + embeddings)

Setup

1. Create a virtualenv and install dependencies:

python -m venv .venv
source .venv/bin/activate
pip install streamlit pandas plotly psycopg2-binary asyncpg anthropic httpx

2. Configure environment variables:

export DATABASE_URL="postgresql://localhost/memory_system"
export ANTHROPIC_API_KEY="your-anthropic-key"
export VOYAGE_API_KEY="your-voyage-key"

3. Initialize the database:

createdb memory_system
psql "$DATABASE_URL" -f schema.sql

Run the app

streamlit run app.py

Common workflows

Generate embeddings

Populates events.content_embedding for semantic search:

python generate_embeddings.py

Generate a daily follow-up plan (CLI)

python followup_agent.py

Notes

• app.py uses query_v2.py and followup_agent.py via subprocess.
• DATABASE_URL defaults to postgresql://localhost/memory_system if unset.
• If you skip embeddings, semantic search will be limited.